CircRNAs derived from the tyrosine phosphatase PTPN22 impact chemosensitivity in ALK-positive T-cell lymphomas

Abstract

Anaplastic large cell lymphoma (ALCL) is a subtype of T-cell non-Hodgkin lymphoma (NHL), classified into ALK(+) and ALK(–) subtypes, based on translocations of the ALK gene. ALK(+) ALCL have a favorable prognosis with polychemotherapy, yet some patients develop early chemoresistance, leading to treatment failure. The molecular mechanisms underlying this resistance remain poorly defined. Circular RNAs (circRNAs) have recently emerged as regulators of drug resistance in cancer, but their role in T-NHLs is largely unexplored. A comprehensive analysis of circRNA expression was performed here in primary ALK(+) ALCL biopsies. RNA-Seq identified 12 circRNAs associated with early relapse, including isoforms from the PTPN22 gene (circPTPN22), significantly upregulated in relapsed patients. Functional analyses showed that the ALK/STAT3 signaling pathway regulates circPTPN22 expression, linking oncogenic signaling to circRNA regulation. It was also found that circPTPN22 isoforms modulate responses to chemotherapy by regulating the arginine methyltransferase CARM1, a key enzyme involved in transcriptional regulation. Loss of CARM1 expression promoted drug tolerance in chemosensitive ALK(+) lymphoma cells. These findings identify the circPTPN22/CARM1 axis as a regulator of chemosensitivity and propose CARM1 inhibition as a potential strategy to overcome chemoresistance in patients with ALK(+) ALCL.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-54978-z.

Introduction

Anaplastic Large Cell Lymphoma (ALCL) is a distinct subtype of T-cell non-Hodgkin lymphoma (NHL) accounting for approximately 3% of adult NHLs and 10 % of childhood lymphomas. All ALCLs are characterized by high CD30 expression¹, which is induced upon lymphocyte activation by antigens or mitogens². ALCL have a wide histological spectrum, including common type, small-cell, lymphohistiocytic, and sarcomatoid variants. The common pattern is the most frequent, accounting for 60% of the cases, followed by the small-cell variant (5–10%). Overall, the prognosis of ALCL expressing the anaplastic lymphoma kinase (ALK) is significantly better than the ALK(–) ALCL. However, although all small-cell variant cases have been shown to be ALK(+), the prognosis is very poor for this subgroup³.

In contrast to adult ALCL, 90% of pediatric ALCL cases are commonly defined by expression of the ALK protein, due to chromosomal translocations involving the ALK gene^4–6. 70% of ALK(+) ALCL carry the t(2;5)(p23;q35) translocation which fuses the ALK gene, on chromosome 2, to the nucleophosmin 1 (NPM1) gene, on chromosome 5, resulting in upregulated, ligand-independent activation of ALK, a tyrosine kinase receptor^1,6,7. Aberrant expression of the chimeric kinase NPM1::ALK activates multiple oncogenic pathways including the Janus kinase (JAK)-STAT3 pathway which is a critical pathway for the proliferation and survival of ALK(+) ALCL cells⁸. STAT3 promotes epigenetic gene silencing by recruiting the DNA methyltransferase 1 (DNMT1) to the promoter of multiple genes. This results in the silencing of T-cell receptor-related genes, leading to the loss of T-cell identity⁹. The STAT3/DNMT1 axis also contributes to the deregulation of non-coding RNAs with oncogenic potential, such as microRNAs^10–13.

Multiagent CHOP (cyclophosphamide, doxorubicin, vincristine, prednisone)-like chemotherapy (CT) remains the standard of care for patients newly diagnosed with ALK(+) ALCL, resulting in long-term remission in the majority of cases. However, the emergence of chemoresistance has become a major therapeutic challenge, limiting treatment efficacy in a subset of patients. Responders to CT experience prolonged complete remissions lasting several years with no detectable minimal residual disease. Relapse occurs in approximately 30 to 40% of patients and is associated with poor prognosis¹⁴. Understanding the mechanisms of chemoresistance is crucial to the development of targeted therapeutic strategies to improve patient prognosis.

Non-coding RNAs (ncRNAs), in particular microRNAs, are involved in the regulation of ALK(+) ALCL drug resistance¹⁵. In recent years, the landscape of non-coding RNAs has expanded dramatically, revealing intricate regulatory networks that control cellular processes. More recently, circular non-coding RNAs (circRNAs) have emerged as a fascinating class of endogenous RNAs. CircRNAs are single-stranded, covalently closed ncRNAs without 5’-end cap or 3’-end poly(A) tail¹⁶. They are generated from their precursor RNA by non-canonical splicing. CircRNAs with unique ‘head‑to‑tail’ splice junctions, called backsplice junctions (BSJ), have the characteristics of broad expression, high conservation and stability¹⁶. CircRNAs can act as transcription modulators¹⁷, microRNA sponges¹⁸ or protein templates¹⁹, and can be translated into peptides²⁰ in some cases. Accumulating evidence suggests that circRNAs play a pivotal role in the process of cancer development and are potential biomarkers for cancer diagnosis, prognosis, and treatment^16,21. While most studies have focused on solid tumors, B-cell malignancies²² and a few T-cell leukemias^23,24, a single publication in 2022 reported the impact of circKIF4A on the progression of an extra-nodal lymphoma subtype²⁵. As a result, T-cell-derived NHL such as ALCL remain unexplored.

Here, results of total RNA sequencing (RNA-Seq) of biopsies from patients with ALCL ALK(+) at relapse are presented. This disclosed two circRNA significantly upregulated in these tissues, potentially involved in sensitivity to CT.

Results

Identification of PTPN22 derived circRNAs upregulation

Previously generated ribosomal RNA-depleted RNA-Seq data^26,27, from diagnostic biopsies of 39 patients with ALK(+) ALCL (R, n=18, noR, n=21,) were analyzed (Supplemental Table S1). Nine reactive lymph nodes (RLNs) served as controls (Supplemental Table S1). Sequencing yielded an average of 247 million reads per sample and identified 62,039 circRNAs across all chromosomes; chromosomes 1 (9.4%) and 18 (8.3%) being the most enriched (Supplemental Figure S1A and B). Notably, 5,290 circRNAs arised from alternative splicing (multiple circRNAs originating from the same parental gene). These alternative events were more abundant in R patients (Supplemental Figure S1 C). Using a minimum cutoff of one read in all R samples, 1,177 circRNAs were identified (Supplemental Figure S2A and B). Heatmap analysis revealed distinct expression patterns between ALCL samples and controls (Supplemental Figure S2 C), most circRNAs originating from exonic regions (96%) (Supplemental Figure S2 D).

Differential expression analysis comparing ALCL with RLN revealed that 279 (23.7%) and 281 (23.8%) circRNAs were upregulated, while 133 (11.3%) and 149 (12.6%) were downregulated in R-ALCL and noR-ALCL, respectively (Fig. 1A). Among these, 12 circRNAs (log2 FoldChange ≥1.5 or ≤–1.5; adjusted P < 0.05) were significantly differentially expressed between R and noR samples (Fig. 1A and Supplemental Figure S3). Specifically, three circRNAs (CDR1, TNFRSF21, BNC2) were downregulated, while nine (EIF4E3, SEL1L3, ARHGEF12, three PTPN22 isoforms, and three TBC1D31 isoforms) were upregulated in R samples (Fig. 1A and B).

Fig. 1.

(A) Differentially expressed circRNA, (B) expression (normalized counts) of 12 differentially expressed circRNAs and (C) quantification of circRNAs using k-mer counts from RNA sequencing datasets generated from healthy tissue/reactive lymph nodes (RLN) (n=9), no relapse (noR, n=21) or relapsed (R, n=18) ALK(+) ALCL samples (n=39).

Given the association of PTPN22 with clinical outcomes that we previously reported²⁶, and the observation that three circRNAs derived from this gene were among the most overexpressed in R patients (Fig. 1A and 1B), focus was brought on these isoforms (circPTPN22#A, #B, and #C). Using RNA-Seq data from ALK(+) ALCL biopsies and control lymph nodes, the dataset was reanalyzed with Transidepia tool, to quantify linear PTPN22 mRNA and circPTPN22#A, #B, and #C expression^28,29 as previously described³⁰. Traditional alignment-based methods often fail to accurately detect non-linear BSJ, limiting circRNA quantification. To overcome this, an alignment-free approach was used that directly analyzes raw sequencing reads for k-mer motifs specific to circular junctions, enhancing both sensitivity and accuracy, particularly for unannotated circRNAs. To specifically differentiate between linear and circular PTPN22 transcripts, 31-nt k-mers were designed, targeting the linear exon-exon junction (linE3/E4) absent in the three circPTPN22 isoforms, as well as k-mers specific to the BSJ of each circRNA: BSJ_E11/E8 for circPTPN22#A, BSJ_E18/E15 for circPTPN22#B, and BSJ_E14/E8 for circPTPN22#C (Supplemental Table S2 and Supplemental Figure S4). This reanalysis confirmed a significant upregulation of both linear PTPN22 mRNA and the three circular isoforms in R ALK(+) ALCL biopsies compared to noR and RLNs (Fig. 1C). Next, the existence of the three circPTPN22 isoforms was verified in the circBase and circBank databases (circPTPN22#A: hsa_circ_0000111/hsa_circPTPN22_002; circPTPN22#B: hsa_circ_0000110/hsa_circPTPN22_005; circPTPN22#C: hsa_circ_0110529/hsa_circPTPN22_003), confirming their exonic nature with splice lengths of 375, 356, and 1354 nucleotides, respectively. CircPTPN22#A originates from exons 8–11 (circPTPN22#A(8,9,10,11)), circPTPN22#B from exons 15–18 (circPTPN22#B(15,16,17,18)), and circPTPN22#C from exons 8–14 (circPTPN22#C(8,9,10,11,12,13,14)) of the PTPN22 gene (Supplemental Figure S4). RNase R and actinomycin D treatment in SU-DHL-1 cells confirmed their greater stability compared to linear transcripts (Supplemental Figure S5 A and B). Subcellular fractionation showed that these circRNAs are predominantly cytoplasmic (Supplemental Figure S5C). Finally, Sanger sequencing validated their BSJ and exon composition (Supplemental Figure S6 and Supplemental Table S2), confirming that they are bona fide circular RNAs expressed in NPM1::ALK(+) lymphoma cells.

Diagnostic significance of circRNA-PTPN22 in ALK(+) ALCL primary biopsies

Previous studies have highlighted the potential of circRNAs as diagnostic tools. The three circPTPN22 isoforms upregulated in patients with ALK(+) ALCL refractory to CT, more stable than linear RNA, could be potential biomarkers of response to CT. Given the poor prognosis and resistance to CT associated with small variant ALCL, RNA-Seq data was analyzed from 39 ALK(+) ALCL primary biopsies including 16 small variant samples (Supplemental Table S1). The latter did indeed display PTPN22 upregulation in relapsing samples particularly in the most aggressive variants, i.e., small cell variants (Fig. 2A). In addition, k-mer analysis³¹ showed increased expression of the three circPTPN22 isoforms in small variant ALCL compared to the common type (Fig. 2B), linking PTPN22 RNAs to drug resistance and disease aggressiveness. A Receiver Operating Characteristic (ROC) analysis provided areas under the curve (AUC) of 0.799 (95% confidence interval [CI], 0.635 - 0.941; P<0;028), 0.819 (95% CI, 0.675 - 0.964 , P<0;0008) and 0.811 (95% CI, 0.669 - 0.954, P <0;001) for circPTPN22#A, #B, and #C respectively (Fig. 2C), indicating good discriminatory ability (AUC ≥ 0.8)³² for circPTPN22#B and #C. To assess the expression of circPTPN22 isoforms, we first evaluated the relative levels of each isoform in ALK(+) ALCL cell lines. Given that circPTPN22#A displayed consistently low expression in both primary biopsies (Fig. 1B) and all cell lines (Fig. 3A), it was excluded from further analyses to ensure the robustness of our results.

Fig. 2.

Using RNA sequencing datasets (A) differential PTPN22 mRNA expression analyses (log2 of the Transcript Count Per Million (TPMs) +1) in ALK(+) ALCL (n=39) primary biopsies including no relapse (noR, n=21), relapsed (R, n=18) samples and common (green, n=23) or small (red, n=16) morphological variants. (B) Three circPTPN22 isoforms (circPTPN22#A, #B and #C) quantified using k-mer counts from RNA sequencing datasets generated from common and small morphological variants. (C) ROC curves constructed to discriminate the 3 circPTPN22 isoforms (circPTPN22#A, #B and #C) in relapsing tissue from not relapsing tissue. AUC = area under the curve. Data are means±SEM (bars), *P<0.05; ****P<0.0001; unpaired two-tailed Student’s t-test with Welch correction.

Fig. 3.

Quantitative real-time PCR (qRT-PCR) analysis (A) of circPTPN22 isoforms (circPTPN22#A, #B and #C) in the 5 ALK(+) ALCL cell lines (KARPAS-299, SU-DHL-1, SUPM2, Pio and COST) bearing the t(2;5)(p23;q35) translocation and (B) of PTPN22 mRNA in human CD3+ lymphocytes activated (CD3_Act, n=5) or not (CD3_noAct, n=5), activated CD3⁺ lymphocytes overexpressing NPM1::ALK cDNA (CD3_Act ALK(+)) (31), and in the 5 ALK(+) ALCL cell lines. MLN51 was used as internal control and the relative ratios of RNA expression were expressed as 2^−ΔCt relative. Data are means±SEM (bars), **P<0.01; ***P<0.001; unpaired two-tailed Student’s t-test with Welch correction.

NPM1::ALK tyrosine kinase activity is responsible for the aberrant linear and circRNA PTPN22 expression in ALK(+) lymphoma cells

Divergent primer pairs spanning the BSJ were designed to specifically amplify PTPN22 circRNAs but not corresponding linear transcripts, the latter being amplified by convergent primers (Supplemental Table S2 and Supplemental Figure S4). RT-qPCR analysis showed a significant increase in PTPN22 mRNA expression following anti-CD3/anti-CD28 stimulation in healthy CD3⁺ lymphocytes (n = 5) compared to non-activated CD3⁺ lymphocytes (n = 5) (Fig. 3B). Interestingly, in activated ALK(+) CD3⁺ cells (n = 2, generated via NPM1::ALK cDNA overexpression³³) and in ALK(+) ALCL cell lines (SU-DHL-1, KARPAS-299, COST, SUPM2, Pio), PTPN22 mRNA levels were markedly lower than in activated normal CD3⁺ cells (Fig. 3B). Expression levels of the three circPTPN22 isoforms varied among ALK(+) cell lines (Fig. 3A). Consistent with these findings, linear PTPN22 transcript levels were significantly downregulated in all ALK(+) lymphoma cells compared to normal activated CD3⁺ lymphocytes (n=5, Fig. 2A). NPM1::ALK activity promotes gene silencing via STAT3-dependent recruitment of DNMT1 to gene promoters in ALK(+) lymphoma cells, which led to examine whether NPM1::ALK and STAT3 influence PTPN22 and its circRNA isoforms. NPM1::ALK and STAT3 were silenced in the COST and SUPM2 cell lines, using siRNAs. Efficient knockdown was confirmed by RT-qPCR (Fig. 4A) and Western blotting (Fig. 4B) and was associated with a low impact on apoptosis (Supplemental Figure S7A). Silencing either NPM1::ALK or STAT3 resulted in a significant upregulation of both linear and circular PTPN22 transcripts in both cell lines (Fig. 4C and D). The role of ALK kinase activity in regulating PTPN22 expression was confirmed by treating COST and SUPM2 cells with crizotinib, which effectively inhibited ALK kinase activity (Fig. 4E) and also led to a decrease in STAT3 activation (p-STAT3 protein levels) (Fig. 4E). Importantly, crizotinib treatment increased PTPN22 expression at both the protein (Fig. 4E) and mRNA (Supplemental Figure S7B) levels, and induced the expression of circPTPN22#B and circPTPN22#C (Supplemental Figure S7B). Crizotinib had minimal impact on apoptosis, as assessed by Annexin V staining (Supplemental Figure S7C). The results show that NPM1::ALK kinase activity and its downstream effector STAT3 can negatively regulate the expression of PTPN22 mRNA and circular isoforms in ALK(+) ALCL cells.

Fig. 4.

ALK and STAT3 mRNA and protein analysis using (A) quantitative real-time PCR (qRT-PCR) and (B) Western blot analysis respectively in NPM1::ALK(+) ALCL cell lines, COST and SUPM2, transfected with control siRNA (si-CTL), si-ALK (siRNA directed against ALK mRNA) or si-STAT3 (siRNA directed against STAT3 mRNA). Vinculin protein was used as an internal control to ensure equal loading. PTPN22 mRNA and circPTPN22#A, #B and #C expression were evaluated by qRT-PCR analysis after si-ALK or si-STAT3 knockdown in COST (C) or SUPM2 (D). MLN51 was used as an internal control and the relative ratios of mRNA and circRNA expression were expressed as 2^−ΔΔCt relative to those in cells transfected with si-CTL. qRT-PCR experiments were performed at least three times. Data are expressed as means±SEM (bars), *P<0.05; ***P<0.001; unpaired two-tailed Student t-test with Welch correction. (E) ALK, pALK, STAT3, pSTAT3 and PTPN22 protein analysis respectively in COST and SUPM2, treated (Crizo) or not (NT) with the ALK tyrosine kinase inhibitor, crizotinib. Actin protein was used as an internal control to ensure equal loading. All experiments were performed at least in triplicate. Blots from three representative experiments are shown.

CARM1 is downregulated in ALK(+) lymphoma cells after silencing of circPTPN22#C

siRNAs targeting the BSJ of circPTPN22#B, circPTPN22#C, and linear PTPN22 RNA were designed and efficiently knocked down these molecules in ALK(+) SU-DHL-1 cells (Supplemental Figures S8A to C). As shown in Fig. 5A, depletion of circRNA-PTPN22#B or circRNA-PTPN22#C at 48h post-transfection, either individually or in combination, did not affect either PTPN22 protein expression nor cell proliferation (Supplemental Figure S8D). Notably, using Annexin V-Pacific Blue/IP staining, it was shown that specific knockdown of PTPN22 mRNA did not affect apoptosis (Supplemental Figure S8E). RNA-Seq analysis of SU-DHL-1 cells transfected with siRNA targeting circRNA-PTPN22#B (Supplemental Figure S8F), circRNA-PTPN22#C (Fig. 5B), both (Supplemental Figure S8F), linear PTPN22 (Fig. 5B), or a control siRNA (siCTL) showed differential gene expression (DEGs, adjusted p-value (padj) < 0.05, log2 fold change ≥ 1 or ≤ -1) for respectively 30, 224, 97 and 122 genes (Supplemental Tables S3–S6) compared to controls, as shown in Volcano plots (Fig. 5B and Supplemental Figure S8F). Among these genes, CARM1 emerged as the most significantly downregulated (padj: 1.79×10⁻⁴⁹; log2FC: -1.064), mostly after silencing circPTPN22#C (Fig. 5C and D). CARM1 expression was validated using RT-qPCR in the same cohort, revealing a significant increase in patients with chemoresistant disease (Fig. 5E). It should be noted that this increase correlates with the high PTPN22 expression that we previously reported in resistant patients²⁶. To further validate CARM1 expression in tumor cells, we first analyzed a bulk RNA-seq dataset generated from ALK(+) ALCL primary biopsies (n = 57). CARM1 expression was positively correlated with tumor cell markers (ALK and CD30) whereas weaker or no correlation was observed with markers of the tumor microenvironment (TME) populations, such as CD19(+) B lymphocytes, αSMA(+) myofibroblasts, CD31(+) endothelial cells, and PDGFRA(+) fibroblasts (Supplemental S9A). To renforce this observation, three regions of interest (ROIs) from two patient primary biopsies were analysed using GeoMx spatial transcriptomic technology (Supplemental S9B). A positive Pearson correlation was also observed between CARM1 expression and CD30/ALK(+) tumor cells (Supplemental S9C), whereas correlations with (CD30(-) TME populations were weak or negative (Supplemental S9D). In addition, given the role of microRNA networks in regulating CARM1, we explored the possibility of microRNA-mediated regulation within this axis. However, our analyses detected no evidence of microRNA sponging or interaction (see Supplementary text and see Supplementary Figure S10). This suggests that circPTPN22#C probably modulates CARM1 expression via microRNA-independent mechanisms in ALK(+) lymphoma cells.

Fig. 5.

(A) Western blot analysis of PTPN22 in SU-DHL-1 cells transfected with control siRNA (si-CTL), si-PTPN22#lin2 (siRNA directed against PTPN22 mRNA), si-circPTPN22-B#B1 (siRNA directed against circPTPN22#B), si-circPTPN22-C#C2 (siRNA directed against circPTPN22#C) or both (si-circPTPN22-B#B1 + si-circPTPN22-C#C2). (B) Identification of differentially expressed genes from the RNA-seq data of SU-DHL-1 cells transfected with si-CTL, si-PTPN22 or the combination si-circPTPN22-B#B1 and si-circPTPN22-C#C2. Volcano plots showing CARM1 as one of the significant downregulated genes upon double si-circPTPN22-B#B1 and si-circPTPN22#lin2 knock down. CARM1 mRNA (C) and protein (D) quantification by quantitative real-time PCR (qRT-PCR) or Western blot analysis respectively. MLN51 was used as an internal control and the relative ratios of mRNA expression were expressed as 2^−ΔΔCt relative to those in cells transfected with si-CTL. Experiments were performed at least in triplicate. The α-tubulin or GAPDH proteins were used as internal controls to ensure equal loading. PTPN22 or CARM1 levels were quantified by densitometry, normalized to the control samples (αtubuline or GAPDH signals), and expressed relative to the control (CTL). Blots from three representative experiments are shown. In ALK(+) ALCL (n=39) primary biopsies including (E) relapsed (R, n=19) or not relapsed (noR, n=21) samples differential CARM1 mRNA expression was analyzed using qRT-PCR. MLN51 was used as an internal control. Data represent means±SEM (bars), *P<0.05; unpaired two-tailed Student t-test with Welch correction.

CARM1 silencing alters the chemosensitivity of ALK(+) lymphoma cells in vitro

CARM1 inhibition has been reported to activate the p53 signaling pathway³⁴, while TP53 gene loss occurs more frequently in ALK(-) ALCL compared to ALK(+) ALCL³⁵. Whole exome sequencing, performed on 57 primary biopsies and NPM1::ALK(+) cell lines disclosed that wild-type TP53 was preserved in 54 out of 57 biopsies and in the Pio and SUPM2 cell lines (Supplemental Table S8). Silencing of CARM1 expression achieved using the CRISPR-Cas9 system in the Pio and SU-DHL-1 cell lines (Fig. 6A and Supplemental Figure S11A). The functional impact of CARM1 silencing was assessed by measuring the arginine dimethylation (me2a) levels of BRG1-associated factor (BAF155) and Poly(A)-binding protein 1 (PABP1), both direct substrates of CARM1. In CARM1-deficient cells, dimethylation of BAF155 and PABP1 was markedly decreased (Fig. 6B and Supplementary Figure S11B). However, analyses performed in Pio KO clones revealed no detectable effects on apoptosis (Supplementary Figure S11C) or cell cycle distribution (Supplementary Figure S11D). In a mouse xenograft model (Fig. 6C and Supplemental Figure S12), CARM1 KO was associated with reduced tumor growth compared to controls. Parental Pio cells and the two stable CARM1 KO clones (Δ CARM1 clone#1 and #2) were treated with CHO chemotherapy. CARM1 KO significantly increased resistance to CHO treatment compared to parental wild type cells (Fig. 6D). It is interesting to note that CARM1 protein levels were significantly lower in cells that had become resistant (SUPM2 CHO) compared to sensitive parental cells (Fig. 6E). It should be noted that both linear PTPN22 mRNA and circPTPN22 levels were reduced in CHO-resistant cells compared with their sensitive parental counterparts (Supplemental Figure S13A), indicating that downregulation of these transcripts accompanies the acquisition and maintenance of chemoresistance. Importantly, once resistance is established, direct comparison of ALK and STAT3 signaling revealed no significant differences in total or phosphorylated ALK and STAT3 between sensitive and resistant cells (Supplemental Figure S13B), demonstrating that the ALK/STAT3 pathway is not further activated in the resistant state.

Fig. 6.

Western blot analysis in two independent knockout clones following CRISPR/Cas9-mediated silencing or not (WT) in NPM1::ALK/TP53(+) Pio cells compared to the non-targeted control sgRNA (ΔCTL, bulk or cloned cells) . (A) CARM1 expression and (B) dimethylation levels of BAF155 and PABP1. The GAPDH protein was used as an internal control to ensure equal loading. Blots from three representative experiments are shown. (C) Cells were injected s.c. into the flank of 8 NOD/SCID mice. Tumor volume was assessed over time by caliper measurements and expressed as mean ± SEM (bars). *P<0.05, using unpaired 2-tailed Student t test. (D) Apoptosis analysis after a combination of cyclophosphamide, doxorubicin, and vincristine (CHO), mimicking CHOP treatment or not (NT) using AnnexinV-Pacific Blue/ PI (flow cytometry). In ALK/TP53(+) SUPM2 resistant to cyclophosphamide, doxorubicin, and vincristine (SUPM2 CHO) compared to sensitive parental cells (WT) treated with TP-064, a selective inhibitor of CARM1. (E) Quantitative real-time PCR (qRT-PCR) analysis of CARM1 mRNA. MLN51 and ABL were used as internal controls and the relative ratios of mRNA expression were expressed as 2^−ΔCt relative. Experiments were performed at least IN triplicate. Data are expressed as means±SEM (bars), *P<0.05, **P<0.01, ***P<0.001; unpaired two-tailed Student t-test with Welch correction.

These results suggest that CARM1 downregulation promotes drug tolerance mechanisms that enhance the survival of resistant ALK(+) lymphoma cells under chemotherapeutic stress.

Discussion

By performing total RNA-Seq analysis on primary biopsies of ALK(+) ALCL, two circRNAs derived from the phosphatase PTPN22 gene (circPTPN22) were found here to be significantly upregulated in tissues from patients with early relapse after CT. It was further demonstrated that ALK and STAT3 modulate circPTPN22 expression, which in turn down modulates the CARM1 arginine methyltransferase expression. Loss-of-function experiments showed that CARM1 deficiency reduces drug tolerance in ALK(+) lymphoma cells. These findings suggest that the circPTPN22-CARM1 axis plays a critical role in regulating chemosensitivity in ALK(+) lymphoma cells.

Protein tyrosine phosphatases (PTPNs) are increasingly recognized as key regulators of oncogenic signaling and drug resistance in ALCL, which is mainly associated with the oncogenic ALK. Downregulation of PTPN1 and PTPN2 leads to hyperactivation of the MEK, SHP2, and JAK/STAT pathways, contributing to resistance against ALK tyrosine kinase inhibitors³⁶. Here, it is shown that PTPN22 is overexpressed in patients with ALK(+) ALCL who relapse early, highlighting the potential role of this phosphatase in resistance to chemotherapy²⁶. The STAT3 signaling pathway plays a central role in ALK(+) ALCL by promoting proliferation and survival of the tumor cells^37,38. It also induces DNMT1, which mediates the methylation-dependent silencing of key T cell receptor (TCR) signaling components and T-cell identity genes, including ZAP70, CD3ε, LAT, and SLP76^27,39–41, explaining the frequent loss of CD3/TCR expression in over 75% of ALCL cases. The ALK/STAT3/DNMT1 axis is demonstrated to be implicated in the silencing of microRNAs involved in treatment resistance, highlighting the role of non-coding RNAs in response to therapy^13,42–45.

Previous studies have shown that ALK signaling represses PTPN1 via STAT3³⁶. Here, it is also demonstrated that the ALK/STAT3 axis suppresses PTPN22 expression as well. Notably, PTPN22 expression was lower in ALK(+) ALCL cells compared to normal T-cells, consistent with its known role as a regulator of TCR signaling. This loss, which is not linked to DNMT1 activity (data not shown), may facilitate persistent oncogenic signaling, contributing to tumour growth and chemoresistance. Interestingly, PTPN22 expression was overall higher in ALK(+) tumor-infiltrated lymph nodes than in reactive lymph nodes. This might be explained by PTPN22 upregulation in non-tumor immune cells within the microenvironment, where it could dampen TCR signaling and promote local immunosuppression. Together, these findings suggest a dual role for PTPN22. Its downregulation in tumor cells may enhance oncogenic signaling, while its upregulation in surrounding immune cells could create a permissive environment for tumor immune escape.

Variants of PTPN22, in particular the R620W polymorphism, have been associated with increased susceptibility to autoimmune diseases such as rheumatoid arthritis (RA), systemic lupus erythematosus (SLE), and type 1 diabetes⁴⁶. Emerging evidence also involved PTPN22-derived circRNAs in disease processes. For example, circPTPN22-E15/E8 is downregulated in PBMC from patients with SLE and RA compared to healthy cells, suggesting its potential as a biomarker⁴⁷. In solid tumors, overexpression of hsa_circ_0000110 (circPTPN22#B) and hsa_circ_0110529 (circPTPN22#C) has been reported with higher levels in the tumor than the adjacent normal tissue, supporting their use as diagnostic and prognostic indicators⁴⁸. In addition, elevated levels of circPTPN22#C have been detected in patients with a subtype of liver cancer, supporting its potential as a diagnostic biomarker for patient stratification and guiding therapeutic decisions⁴⁹. In the present study, circPTPN22#B and circPTPN22#C were also expressed in non-NHL cells, which may limit their diagnostic specificity. However, their elevated expression in ALK(+) ALCL specifically correlates with early relapse, highlighting their potential utility as prognostic biomarkers in this aggressive NHL subtype.

CARM1 (PRMT4) is a key arginine methyltransferase involved in gene regulation and chromatin remodeling. It mediates transcriptional activation via H3R17 dimethylation⁵⁰, and methylates non-histone substrates such as BAF155 and PABP1⁵¹,⁵². Interestingly, while CARM1 deletion impairs tumor cell proliferation in vivo, its expression is markedly reduced in relapsed ALK(+) ALCL patient samples, where PTPN22 is concomitantly upregulated. One plausible explanation is that CARM1 supports tumor growth in primary settings by promoting angiogenesis or metabolic adaptation⁵³. However, under the selective pressures of therapy, CARM1 activity may become dispensable, as alternative survival pathways could compensate its loss. In this context, increased PTPN22 expression might contribute to maintaining tumor cell viability through mechanisms independent of proliferation, such as immune evasion or metabolic adaptation. This adaptive shift aligns with the epigenetic plasticity observed in cancer progression, where tumor cells acquire resistance traits despite reduced proliferative capacity⁵⁴. Together, these observations indicate that circPTPN22 likely exerts its effects independently of microRNAs in ALK(+) ALCL, supporting a model in which circPTPN22 regulates CARM1 through a direct mechanism, such as interaction with proteins or modulation of transcript stability, rather than via microRNA sponging.

The PTPN22 high/ CARM1 low phenotype may thus define a relapse-associated tumor cell state marked by slower growth, enhanced adaptation to a hostile microenvironment and increased capacity for long-term persistence. Such a state may underlie minimal residual disease, maintaining tumor cells in a quiescent or low-proliferative state that is more resistant to conventional chemotherapy. These findings underscore the plasticity of ALK(+) ALCL and support the notion that relapse is driven not solely by increased proliferation, but also by phenotypic shifts that favor survival under therapeutic pressure. These observations are consistent with the concept of drug-tolerant persister cells, which have been increasingly recognized as contributors to chemoresistance across multiple cancer types^55,56. Targeting these compensatory survival pathways that sustain this drug-tolerant state may offer promising strategies to prevent or overcome relapse in ALK(+) ALCL.

Collectively, these findings identify circPTPN22 isoforms and CARM1 as novel actors of chemoresistance in ALK(+) ALCL highlighting the mechanisms underlying tumor cell adaptation and persistence.

Methods

Tumoral and normal samples

A NPM1::ALK(+) ALCL patient cohort at diagnosis was established between 1994 and 2009 (PAIR T lymphoma project). Among them, 18 patients (R) relapsed after CT within a median of 7 ± 3 months following diagnosis, while 21 remained in remission (no relapse, noR) with a median follow-up of 58 months (Supplemental Table S1). Malignant cells accounted for at least 50% of lymph node involvement. Frozen tumor samples were collected with informed consent from the CHU de Toulouse tumor tissue bank, in accordance with institutional review board–approved protocols (DC 2009-989) and the Declaration of Helsinki. Clinicopathological characteristics of the cohort have been published previously²⁶. Nine lymph nodes from patients with reactive non-malignant disease (RLN), used as control tissue samples were provided by the CRB-Cancer of Bordeaux University Hospital (BRIF BB-0033-00036) member of the Cancéropôle Grand Sud-Ouest network.

Cell lines and drugs

The five ALK(+) ALCL cell lines, KARPAS-299, SU-DHL-1, SUPM2, Pio and COST, all carrying the t(2;5)(p23;q35) translocation, were obtained from the DSMZ (German Collection of Microorganisms and Cell Culture) or established locally⁵⁷. Cells were cultured in RPMI 1640 medium, (GlutaMAX™ supplement, Gibco™, Thermo Fisher, Waltham, MA) supplemented with 20% FBS and 1mM of sodium pyruvate. The chemoresistant SUPM2 CHO cell line was provided by C. Gambacorti-Passerini and L. Mologni (Department of Medicine and Surgery, University of Milano-Bicocca, Monza, Italy) and was cultured in RPMI-1640 supplemented with 10% FBS, 2 µM cyclophosphamide, 20 nM doxorubicin; 0.5 nM vincristine, constituting the CHO treatment⁵⁸. All cells were maintained at 37°C in a humidified atmosphere containing 5% CO₂.

Crizotinib was synthesized and purchased from @rtMolecule (Poitiers, France). SUPM2 and COST cells were treated with 300nM and 200nM of crizotinib, respectively. SUPM2 CHO and SUPM2 cells were seeded at 150,000 cells/mL and treated with 25 µM or 10 µM TP-064 (MedChemExpress, Monmouth Junction, NJ) for 72h.

circRNA validation by PCR and sanger sequencing

To validate the BSJ and the complete sequence of circRNA, PCR products were generated using the FastStart Taq DNA Polymerase (Roche, Basel, Switzerland) and submitted for Sanger sequencing by Eurofins Genomics GmbH (Ebersberg, Germany). Circularity and exonuclease resistance were assessed by treating 1 µg of total RNA with 2U RNaseR (Lucigen, Teddington, UK), for 30 minutes at 37°C. Stability was verified by treating cells with 1nM of actinomycin D (Sigma, St Louis, MO) for 12 hours. Treatments were validated by RT-qPCR to assess circRNA and mRNA levels. Target relative expression was assessed by the 2^-ΔCt method. Sequences of all primer pairs are provided in the supplementary data section (Supplemental Table S2).

Cas9 CARM1 knockout and treatment

Three million NPM1::ALK(+) cells were transfected with the RNP/Cas9 complex using the 4D Nucleofector Amaxa technology (Lonza) (using IDT pre-designed guides: Hs.Cas9.CARM1.1.AA AACACCGACACGGTAGCGCA and Hs.Cas9.CARM1.1.AI GAGGACGCTGTTGCAGCCCA) and the Cas9 protein (2:1 ratio) using P3 Primary Cell 4D X Kit L (program CA-137). One week after transfection, cells were cloned at 1 cell per well in 96 well plates. Cells were seeded at 2.5x10⁵ cells/mL. Twenty-four hours later, cells were treated with 20µM cyclophosphamide, 200 nM doxorubicin and 5 nM vincristine, constituting the CHO treatment.

Xenograft tumour assay

Mice were housed under pathogen-free conditions in an animal room at constant temperature (20°C-22°C), with a 12-hour light/dark cycle and free access to food and water. 3 × 10⁶ ALK(+) cells were injected subcutaneously into one flank of 5-week-old female NOD/SCID gamma (NSG) mice (Janvier Labs) under general anesthesia by inhalation of 3% isofluorane. Mouse body weights are measured 1 time a week, and tumour volumes were measured 3 times a week with calipers, using the following formula: length × width² × π /6. At the end of the experiment, mice (4 per group) were humanely sacrificed by cervical dislocation. All animal procedures were performed following the principle guidelines of INSERM. Our protocol was approved by the Midi-Pyrénées Ethics Committee on Animal Experimentation and conducted in accordance with institutional guidelines and the Directive 2010/63/EU under protocol number DAP-APAFiS-2021100714596279-, CEEA122 agreement: 2021100714596279_v9). All experiments were conducted in compliance with all relevant guidelines and regulations, including the Animal Research: Reporting of In Vivo Experiments (ARRIVE) guidelines.

PBMC isolation

Peripheral blood mononuclear cells (PBMCs) from healthy donors were used as control lymphoid cells. Human peripheral blood samples were obtained from anonymous male or female healthy adult volunteers who provided informed consent in accordance with the local research ethics committee of the Etablissement Français du Sang (Toulouse, France, agreement 21PLER2021-007). PBMCs were isolated from whole blood samples by Ficoll density-gradient centrifugation (Pancoll human, PanBiotech GMBH, Aidenbach, Germany) with SepMateTM-50 tubes (StemCell technologies, Vancouver, Canada), from which T-cells were sorted (Pan T cell isolation kit human from Miltenyi Biotec, Paris, France). T-cells were activated with the anti-CD3 (OKT3, Biolegend, San Diego, CA, #317325 RRID: AB_11147370, 5.5µg/mL) and anti-CD28 (eBioscience, Thermo Fisher, #16-0289-81 RRID: AB_468926, 2µg/mL). Cells were cultured in RPMI-1640 GlutaMAX™ supplemented with 20% FBS and 1mM of sodium pyruvate (all from Invitrogen, Waltham, MA). All cells were maintained at 37°C in a humidified atmosphere containing 5% CO₂.

RNA extraction, reverse transcription and quantitative PCR

Total RNA was extracted using the Direct-zol RNA MiniPrep Plus w/ Zymo-Spin IIICG Columns (Capped) kit (Zymo Research, Irvine, CA) according to manufacturer instructions. Total RNAs were reverse transcribed using the ProtoScript® II First Strand cDNA Synthesis Kit (NEB, Thermo Fisher) according to manufacturer recommendations. Expression of circRNA, genes of interest and MLN51 and/or ABL (used as reference genes) was assessed by quantitative PCR using divergent and convergent primers, respectively using SYBR™ Select Master Mix (Applied Biosystems™, Thermo Fisher). The sequences of all primer pairs are provided in the supplementary data section (Supplemental Table S2). Relative expression of targets was assessed by the 2^-ΔΔCt or 2^-ΔCt method.

Protein extraction and Western blot analyses

NPM1::ALK(+) cells were lysed in protein lysis buffer (50 mM Tris-HCl pH 7.4, 1%Triton X-100, 0.1% SDS, 150 mM NaCl, 1 mM EDTA, and 1 mM DTT), supplemented with protease and phosphatase inhibitors (complete protease inhibitor cocktail tablets; Roche #11697498001) and Protease and Phosphatase Inhibitor Cocktail, (Thermo Fisher #78440) during 20 minutes on ice. After removal of the insoluble material (centrifugation for 15 minutes at 16,000g), the total protein concentration was determined using BCA (bicinchoninic acid assay). Proteins were separated on 12% SDS acrylamide gels and transferred to nitrocellulose membranes (Trans-Blot Turbo, Bio-Rad, Hercules, CA) according to the manufacturer protocol. Immunodetections were performed overnight at 4°C, using primary antibodies directed against the protein of interest (dilution 1/1000). Signals from HRP-conjugated secondary antibodies were generated using enhanced chemiluminescence solution (Chemiluminescent Peroxidase Substrate-3 Kit; Millipore, Sigma) or the Clarity Western ECL Substrate (Bio-Rad). Quantification was performed by densitometry using Fiji software. All antibodies are listed in supplementary data (Supplemental Table S2).

Gene/circRNA knockdown by siRNA

Three million NPM1::ALK(+) cells were transfected with 60 pmol or 200 pmol of siRNA by electroporation, using a P3 Primary Cell 4D X Kit L from Lonza Bioscience (Basel, Switzerland). Commercial scramble siRNA from Eurofins was used as negative control. For circRNA downregulation, siRNAs were designed at the level of the BSJ while, for all coding genes, siRNAs were designed at the conventional exon-exon junction. Forty-eight hours after transfection, cells were harvested for RNA and protein extractions. All siRNA sequences are listed in Supplementary data (Supplemental Table S2).

Apoptosis assay

Forty-eight hours after treatment, cells were washed with PBS and stained with Pacific Blue-conjugated Annexin V and propidium iodide (1:1 ratio) (Biolegend). Flow cytometry was performed using a MACS Q10^TM cytometer (Miltenyi Biotec).

Cell proliferation assay

Prolification rate was assessed using violet CellTrace kit (ThermoFischer Scientific #C34557) following the manufacturer’s instructions.

Cell cycle analysis

Cell cycle analysis was performed via an EdU Click-iT kit (Invitrogen, Waltham, MA #C10337) following the manufacturer’s instructions.

RNA sequencing and differential expression analysis

The ribo zero full RNA sequencing (RNAseq) method has previously been published (GSE160123)³³. RNA library preparation from SU-DHL-1 cells with siRNA in triplicate was performed according to manufacturer recommendations (Illumina stranded total RNA prep, ligation with ribo-zero plus, San Diego, CA). Final samples pooled from library preparations were sequenced on Illumina Novaseq6000 with S1-300 cycles cartridge (2x1600M, reads of 150 bases each), corresponding to 2x64M reads per sample after demultiplexing. Small RNA library preparation was performed according to manufacturer recommendations (QIAseq miRNA Library Kit from QIAGEN, Hilden Germany). The final pooled library samples were sequenced on Novaseq 6000 Illumina corresponding to 1x30M base reads per sample after demultiplexing. The reads were aligned to the hg38 human genome using HISAT2. Gene expression was then quantified with featureCounts, and a differential expression analysis was carried out using DESeq2.For RNAseq from primary tumor sample, expression was quantified using Kallisto, a fast pseudo-alignment-based method for transcript quantification from RNA sequencing data⁵⁹. De novo identification of circRNAs was performed according to a standard protocol, using the CIRI2 algorithm with the human reference genome GRCh37/hg19²⁸. circRNA-specific junction reads are non-colinear and therefore typically found within the reads that do not align to the genome. CIRI2 retrieves this “trashed” information to identify and distinguish circular transcripts from linear transcripts.

Reindeer (REad Index for abuNDancE query)²⁹, an efficient indexing method based on k-mers and their associated counts was used to enable rapid quantification of circRNAs of interest in patient and control RNAseq data. After indexing RNA-Seq data, kmer analysis was performed. A sequence of 31 nucleotides was generated at the BSJ of circRNAs (circPTPN22#A(8,9,10,11) /hsa_circ_0000111, circPTPN22#B (15,16,17,18) /hsa_circ_0000110 and circPTPN22#C(8,9,10,11,12,13,14) /hsa_circ_0110529. For PTPN22 mRNA (transcript: ENST00000359785.10, PTPN22-201), the kmer was designed at the classical exon3-4 junction. The abundance of these kmers was quantified in RNAseq data indexed in Transipedia (https://transipedia.org/)²⁸.

Spatial transcriptomic

All experiments and analysis were conducted by the NanoString platform (NanoString Technologies, Seattle, USA) using the GeoMx Digital Spatial Profiler (DSP) technology⁶⁰. Tissue sections were stained using a CD30-specific probe (clone ON6D/C2, catalog #LS-C87916-100, dilution 1:100, channel 647) together with SYTO83 (DAPI) for nuclear counterstaining.

Targeted TP53 DNA sequencing and bioinformatic analyses

TP53 genomic analysis was performed using pan-exon targeted next-generation sequencing of DNA extracted from diagnostic samples. DNA libraries were prepared using the Nextera XT kit (Illumina) and sequenced on a NextSeq (Illumina)²⁶.

Subcellular fractionation

Cells were lysed with 1 mL of 0.1% cold NP-40 diluted in PBS and incubated on ice for 30 minutes before centrifugation at 500g for 3 minutes at +4°C and collection of the cytoplasmic fraction. The nuclear fraction was washed 4 times with ice-cold PBS followed by another wash in 0.1% NP-40. Nuclear and cytosolic RNAs were extracted using a Trizol/chloroform RNA extraction protocol and subcellular localization was assessed by qRT-PCR. GAPDH mRNA and long non-coding MALAT1 were used as cytoplasmic and nuclear RNA controls, respectively. Sequences of all primer pairs are provided in Supplementary data (Supplemental Table S2). Relative expression of the targets was assessed by the 2^-ΔCt method.

Statistical analyses

Data are expressed as mean ± standard error of the mean (SEM) of at least three independent experiments. ROC curves, graphs and statistical analyses were performed using GraphPad Prism 10 (GraphPad Software, Inc., La Jolla, CA). P values were calculated by using unpaired two-tailed Student t-test with Welch correction, and considered statistically significant if lower than 0.5.

Author contributions

E.A., L.B. and F.M. designed the research study. E.A., S.F., C.P., I.D., C.Q., R.F., A.Z. and L.B. performed the experiments. E.A., L.B., S.F., C.Q. and F.M. analyzed the data. L.L. and C.Q. were involved in the diagnosis of ALCL. C.B., L.B. and C.G. performed the bioinformatic analysis. S.P. and M.B. helped with discussions and critical reading. L.B. and F.M. wrote the manuscript. All authors have read and agreed to the published version of the manuscript.

Funding

The study was supported by INSERM, Labex Toucan, Association Eva pour la vie, Féderation Grandir sans Cancer, RACCE Lions Club de Lourdes, Association le sourire de Lucie, le marcheur au grand coeur/Fondation les batisseurs d’étoiles, the French Ministry of Health and the French National Cancer lnstitute (CircOma, PRT-K 2022-184), Fondation ARC et Fondation pour la Recherche Médicale supported by its sponsor Nagui and his production company, Banijay Production (EQU202403018019). All grants were obtained by F.M. E.A. was supported by Labex Toucan/Laboratoire d’Excellence Toulouse Cancer, EUR CARe N°ANR-18-EURE-0003 in the framework of the Programme des Investissements d’Avenir and the association Cassandra. L.B and C.B were supported by Fondation de France. L.B. was supported by Fondation l’Oréal for Women in Science, Prolific Graine de Chercheur, Association Ellye, Fondation ARC and Fondation Toulouse Cancer santé. L.C., S.D. and S.F. were supported by a grant from the French Ministry of Health and the French National Cancer lnstitute (CircOma, PRT-K 2022-184). S.F. was supported by a grant from the Deutsche Forschungsgemeinschaft (DFG, grant no. 439441203). S.F. is a participant in the BIH-Charité Clinician Scientist Program funded by the Charité - Universitätsmedizin Berlin and the Berlin Institute of Health.

Data availability

RNAseq datasets from patients with ALK(+) ALCL, were previously published 26. All other data generated or analysed during this study are available within the article and its supplementary information files. Additional data or materials are available from the corresponding authors upon request: Fabienne Meggetto (fabienne.meggetto@inserm.fr).

Declarations

Competing interest

The authors declare no competing interests.