Work

Common genomic and transcriptomic signatures in Richter transformation highlight druggable vulnerabilities and guide drug repurposing strategies

Year: 2026

Type: article

Abstract: Richter transformation (RT) is a rare and aggressive lymphoma occurring in patients with chronic lymphocytic leukemia (CLL). Despite advancements in understanding the pathogenetic mechanisms driving disease transformation, it remains a challenging malignancy [ 1 , 2 , 3 ]. In recent years, multi-center genomic studies have shed light on the heterogeneity of RT, with no single driver alteration identified [ 4 , 5 , 6 , 7 ]. This complexity underscores the need for a comprehensive molecular profiling through multi-omic approaches, as one-fits-all therapeutic strategies are unlikely to succeed. Such profiling offers a way to capture actionable vulnerabilities, based on experimental approaches that identify and prioritize “recurrent players” across cohorts. To identify novel therapeutic opportunities through a drug repurposing strategy - aimed at accelerating translation from research to the clinics - we investigated paired genomic and transcriptomic data from a cohort of 20 RT patients. Results were then integrated with data from 3 recently published cohorts [ 4 , 5 , 8 ] to derive a RT “blueprint”, and define a hierarchy of consistent disease hallmarks. The convergent features were exploited to design an ad-hoc drug library, functionally validated through a high-throughput screening in RT models (Fig. 1A ). Fig. 1: Characterization of the genomic and transcriptomics landscape of RT led to the identification of a hierarchy of actionable signatures. Full size image A Graphical description of the crucial steps of the study workflow from nucleic acids isolation to definition of common hallmarks, based on data the genomic and transcriptomic data integration. Numbers in brackets indicated of the number of samples for each step. B Oncoprint of pathogenic and likely pathogenic single-nucleotide variants (SNVs) and small insertion/deletion (indels) identified in a cohort of 20 RT patients. Horizontal bar plot (graph on the right, Variants/patient) indicated the number of variants per patient, whereas veritical bar plot (bottom graph, Variants/gene) showed the number of pathogenic/likely pathogenic variants per gene. Genes are listed based on frequency of alteration within the study cohort. In blue are indicated genes carrying novel variants, never described before to be altered in RT. The different types of variants are indicated in a color code: missense (blue), frameshift (green), splicing (fuchsia), nonsense (darkpurple), and inframe-indel (yellow). C – D Venn diagram of detected genomic alterations (SNVs) in four separate cohorts: Turin (green), Klintman’s (blue), Nadeu’s (pink), and Parry’s (yellow) ( C ) and Venn diagram of up- and down-regulated genes of three separate cohorts (in-house cohort, Nadeu’s, Parry’s) ( D ). Numbers inside the diagram indicate the number of variants ( C ) or differentially expressed genes ( D ) detected in each cohort or in common. Colored blocks summarize the significant MSigDB Hallmark enrichment results for each Venn-diagram overlap ( P value < 0.05). For each overlap, genes shared between cohorts are indicated by the corresponding color in the legend, and the enrichment analysis was performed on the gene set defined by that specific overlap. Numbers in the Venn diagrams (and in the corresponding colored blocks) indicate the number of shared genes in each overlap. Enriched terms shown in bold denote terms observed in both the genomic- and transcriptomic- analyses. DNA and RNA were co-extracted from 20 formalin-fixed paraffin-embedded (FFPE) lymph-node biopsies from RT patients collected between 2010 and 2020. Targeted DNA sequencing was performed using TruSight Oncology 500, enabling detection of clinically relevant single-nucleotide variants (SNVs)/indels and copy-number variations (CNVs) within cancer-associated genes. Bulk RNA sequencing was feasible in 14/20 samples and allowed us to identify differential gene expression (DGE) compared to the preceding disease phase, exploiting multiple GEO/EGA CLL datasets ( n = 250). Although being a valid source of material and information for genetic analyses, FFPE samples may present quality issues due to the conservation technique [ 9 , 10 ]. Methodological fidelity was assessed in a matched comparison, using the RT-PDX model RS1316, sequencing both DNA and RNA from fresh material, and from its FFPE counterpart. Variant allele frequencies (VAF) for five known variants in RS1316 [ 11 ] were preserved between conditions (paired t-test not significant), with a Spearman correlation indicating strong concordance (ρ = 0.9; Supplementary Fig. S1A ). For transcriptomics, linear regression of log 10 TPMs between fresh and FFPE material showed strong correlation (R² = 0.74; P < 0.0001; Supplementary Fig. S1B ). In addition, a computational deconvolution analysis using MCPCOUNTER module of TIMER2.0 showed a significant enrichment in the B cell fraction - being the most predominant population - whereas other immune fractions including T, NK cells, and monocytes were almost undetectable, reducing concern that cohort signatures were driven primarily by microenvironmental contamination (Supplementary Fig. S1C ). Targeted genomic profiling highlighted extensive inter-patient heterogeneity, consistent with the view that RT does not depend on a single universal driver alteration [ 8 ]. Variant calling data allowed the detection of 3678 unique variants, 53.9% of which were in coding regions. The identified variants were predominantly synonymous (53.9%) and missense (40.2%), followed by splicing (2.1%), frameshift (1.6%), in-frame indels (1.4%), and nonsense (0.8%; Supplementary Tables S1 , S2 ). Filtering for somatic status and VAF > 0.10 yielded 102 unique coding variants (Supplementary Tables S3 , S4 ), 56 of which were classified as pathogenic/likely pathogenic (C5/C4), while the remaining were variants of unknown significance (C3; Fig. 1B ). TP53 was the most frequently altered gene, mutated in 45% of patients (9/20), in line with prior reports [ 4 , 5 , 6 ]. Other recurrent alterations, present in at least two patients of the cohort, affected ARID2 and RHOA genes (10% each; Fig. 1B ). The remaining detected variants were patient-specific and included genes previously reported as recurrently mutated in RT, such as ATM , BIRC3 , BRAF , NOTCH1 , and XPO1 (Fig. 1B ). CNVs were observed in 14 out of 20 patients, with a total of 82 events affecting 29 genes and more than half (56.4%, 38 gains and 27 losses) were found in at least two patients (Supplementary Fig. S1D and Supplementary Table S5 ). CNVs involving CDK6 , MYC , and BRAF were also detected in our cohort, highlighting their recurrent involvement in RT and further reinforcing their role in disease biology. Enrichment analysis of genes affected by genetic alterations (SNVs/indels and CNVs) highlighted Wnt/β-catenin, JAK-STAT3/MAPK, and NOTCH signaling, alongside DNA damage, cell cycle, and apoptosis, as the most affected pathways (Supplementary Fig. S1E, F ). We next characterized the RT transcriptional profile. Principal component analysis indicated that RT samples clustered together and were distinctly separated from CLL ones (Supplementary Fig. S2A ). RNA-seq data analysis identified 1964 differentially expressed genes between RT and CLL, 1156 of which up-regulated and 808 down-regulated (Supplementary Fig. S2B and Supplementary Table S6 ). Enrichment analysis highlighted up-regulation of genes related to proliferation/cell-cycle progression, DNA damage response and repair, metabolic rewiring, microtubule cytoskeleton organization, and apoptosis regulation. In contrast, the down-regulated signature revealed significant enrichment of RNA metabolism-related terms, including splicing (Supplementary Fig. S2C ). Of note, down-regulation of ribosomal RNA processing and ribonucleoprotein biogenesis was also observed (Supplementary Fig. S2C ), potentially causing cascading effects on cell survival, cell cycle progression, DNA repair, and apoptosis [ 12 ]. To define a robust RT “blueprint”, data from our cohort were then integrated with the ones obtained in independent RT cohorts (3 genomic and 2 transcriptomic) [ 4 , 5 , 6 ], performing a combined genomic-transcriptomic analysis. While gene overlap across cohorts was limited, pathway analyses showed clear convergence, with cohort-specific genes insisting on shared dysregulated programs, including PI3K/AKT/mTOR, IL-6/JAK/STAT3, G2-M checkpoint, mitotic spindle, DNA repair, NOTCH signaling, and apoptosis (Fig. 1C ). In transcriptomics, up-regulated cross-cohort intersections converged on proliferative/stress programs and metabolic adaptation, including glycolysis, ROS, fatty-acid and xenobiotic metabolism (Fig. 1D , left), whereas down-regulated intersections implicated NOTCH signaling, immune-related terms, hypoxia, and UV-response (Fig. 1D , right). Even in the case of transcriptomics, a considerable fraction of mRNAs was cohort specific, but enrichment analyses showed that they largely merged on the same pathways, with overlap across cohorts (Supplementary Figs. S3 – S5 ). Overall, these data allowed us to uncover shared actionable disease hallmarks that are worthy to be experimentally validated, in a translational perspective, through a drug-repurposing-driven approach. To this aim, we combined a pathway-based (shared dysregulated processes with experimental support for their role in RT biology) with a target-based selection (mutated/differentially expressed genes in RT) to catch drugs in phase III/IV of clinical trials or already FDA-approved. Using these criteria, a commercially available drug library, made of 3,585 drugs, was filtered out resulting in 66 compounds (Supplementary Fig. S6 and Supplementary Table S7 ) that were tested in high-throughput screening on RT-PDX models and the U-RT1 cell line, in two consecutive steps. At 24 h (first round), responses were heterogeneous and dose-dependent, with several compounds showing activity only at higher concentrations and with clear inter-model variability. At the highest dose (10 µM), six drugs targeting nucleotide metabolism (Methotrexate, Prelatrexate, and Pemetrexed disodium) and DNA repair (Azathioprine, Cytarabine, and Mercaptopurine) were active exclusively in the U-RT1 cell line, while two drugs targeting the cell cycle (Colchicine and Volasertib) and one targeting protein tyrosine kinases (Ceritinib) showed efficacy in the RT-PDX models but not in U-RT1 (Fig. 2A, B ). Importantly, 7 compounds (Afatinib, Clofarabine, Homoharringtonine, Idarubicin, Midostaurin, Nilotinib, Ponatinib) showed activity across all tested models (Fig. 2B ). Based on a predefined growth-rate threshold (<0.6 at 10 µM for all drugs and 2 µM for neratinib at 24 h), 27/66 compounds were advanced to a second screening round at 48 h (Fig. 2C ; Supplementary Fig. S7A–E ). At this time point, drug effects were stronger and more consistent, showing responses at 1 µM and 0.1 µM, and supporting time- and dose-dependent target engagement rather than mere cytotoxicity (Fig. 2D ). Despite heterogeneity, five compounds (Gemcitabine, Idarubicin, Irinotecan, Homoharringtonine, and Colchicine) were cross-active in RT cells (Fig. 2D ; Supplementary Fig. S7F ). Notably, Irinotecan and Gemcitabine broadened efficacy at 48 h and at lower doses compared with the 24 h assay (Fig. 2D ). Moreover, in both rounds of screening, RT-PDXs exhibited a comparable profile of response, slightly different from the U-RT1 (Fig. 2A–D ), which showed no response to drugs targeting the PI3K pathway, consistent with previously reported insensitivity to Duvelisib (Fig. 2D and Supplementary Fig. S7E ) [ 13 ]. Noteworthy, 16 compounds - 14 of which are FDA-approved - achieved significant reduction in growth rate at 1 µM after 48 h in at least one model, highlighting their translational potential (Supplementary Fig. S7G ). Together, these results show that despite RT heterogeneity, a subset of ready-to-use candidates may be explored in clinical settings (Supplementary Fig. S7G and Supplementary Table S8 ). It is important to underline that all the investigated compounds were tested as single agents: we can speculate that combination strategies are worthy of further investigation, with the potential to enhance efficacy and enable dose-sparing regimens. Fig. 2: High-throughput drug screening reveals actionable vulnerabilities in RT. Full size image A Heatmap showing the drug response profiles of the 66 selected compounds in the first round of screening (24 h) across the models (RS1316, RS9737, RS1050, IP867/17, and U-RT1). Each drug was tested at three different concentrations (10 µM, 1 µM, and 0.1 µM), except for Neratinib, Itraconazole, and Palbociclib, which were tested at 2-0.2-0.02 µM. Color scale represents relative cell viability, with yellow to red denoting decreased cell growth. Drug names are listed on the left, and corresponding pathways they belong to are shown on the right (color-coded). B Venn diagrams illustrating shared and model-specific effective drugs (growth rate <0.6) across models at 24 h at the 10 µM (upper panel) and 1 µM (lower panel) concentration. C Bar plot summarizing the distribution of the 27 drugs selected for second round of screening, based on their target pathways, over the 66 initially tested compounds (each square corresponds to a drug and, if colored, represents a compound selected for secondary round of screening). D Heatmap of the second round of screening at 48 h across all models, using the same conditions as above. Drug names and their corresponding pathways are indicated on the left. In conclusion, these results indicate that, rather than focusing on individual genetic lesions, more promising insights into key relevant disease targets may be achieved by taking a step-back and observing the broader context - focusing on gene categories and pathways. Instead of relying on blind high-throughput drug screening, employing a multi-omics-guided target prioritization allows the identification of the most promising and effective molecules to be exploited for the treatment of RT. In addition, these results support the feasibility of a drug-repurposing strategy that may accelerate the bench-to-bedside transition. Despite the disease models used represent only a small fraction of the potential molecular diversity of RT, the obtained results highlight the presence of common vulnerabilities that can be therapeutically relevant.

Source: Leukemia

DOI: 10.1038/s41375-026-02964-9