The data analyzed were extracted from National Center for Biotechnology Information (NCBI) Gene Expression Omnibus (GEO) datasets GSE75122 (15), GSE137024 (16), and GSE92626 (4) ( Supplementary Table S1 ). To identify differentially expressed genes (DEGs), data were processed as described elsewhere (17, 18) ( Supplementary Figure S1 ). We used the R packages SVA (ComBat function) (19) to adjust combined data for batch effects and RankProd (20) to calculate the rank product (RP), a non-parametric method based on the estimated percentage of false predictions (pfp). Gene Ontology and Reactome pathway analyses were performed using gprofiler (21) and the Reactome database (22), respectively, with a threshold of 0.05 (both). To visualize significantly enriched Reactome pathways [false discovery rate (FDR)< 10%], “makeDendrogram.py” (pyEnrichment: https://github.com/ofedrigo/pyEnrichment) and custom in-house R scripts were used to construct bubble plots using multidimensional scaling to calculate an optimal two-dimensional (2D) arrangement of pathways based on a distance matrix of between-pathway semantic scores. All these analyses were carried out in the R environment. The gene–gene interaction was evaluated with the GeneMANIA tool (23) using Cytoscape v3.9.1.

All participants signed written informed consent forms in accordance with the Declaration of Helsinki, and laboratory protocols were approved by the Institutional Review Board of the University of Perugia. Primary CLL cells were isolated from peripheral blood (PB) with a purity of 93.8% ± 2.7% CD19⁺/CD5⁺, assessed by flow cytometry using anti-human CD45, CD19, CD5, CD11b, and CD3 monoclonal antibodies on 7AAD-negative cells ( Supplementary Table S2 ), and characterized for IGHV and SF3B1 mutational status and the main cytogenetic abnormalities as described elsewhere (24–26). The NOTCH1 mutation (c.7541_7542delCT) allelic burden of CLL cells was determined by digital PCR (ddPCR), as previously described (27), and stratified according to allele mutation frequency as low (from 0.03% to 12%) and high (above 12%), as described elsewhere (5, 28).

Cells were cultured at a density of 2 × 10⁶ cells/ml in complete medium, consisting of RPMI 1640 supplemented with 10% heat-inactivated fetal bovine serum (Gibco), 2 nM L-glutamine, 100 U/ml penicillin, and 100 μg/ml streptomycin (all from Invitrogen), with the following agents: dimethyl sulfoxide (DMSO) as a vehicle, curcumin (Sigma-Aldrich, St. Louis, MO, USA), and/or ABT-199/venetoclax (Selleck Chemicals, Houston, TX, USA). EDTA (0.5 mM) was added to specifically activate NOTCH1 signaling.

Lentiviral packaging was conducted using 293T cells at a confluency of 70%–80%. 293T cells (15 × 10⁴ cells) were transiently transfected with an FG12-based lentiviral vector expressing green fluorescent protein (GFP), containing pNICD WT or pNICD-mutated fragment (c.7541_7542delCT), or is empty, used as a mock control (7.5–10 µg plasmid DNA). Lentiviral vector stocks were generated using pCMVR8.74 (packaging plasmid) and pMD2.G (VSV-G envelope expressing plasmid). All plasmids were obtained from Addgene (Watertown, Massachusetts). FG12 was a gift from David Baltimore (Addgene plasmid #14884) (29), and pCMVR8.74 and pMD2.G were a gift from Didier Trono (Addgene plasmids #22036 and #12259, respectively).

The transfection was performed using T-Pro NTR II transfection reagent (T-Pro Biotechnology) in Opti-MEM medium (Gibco, Thermo Fisher). After collection and filtration of the viral supernatant, all PGA1 cells were subjected to static transduction for 48 h. The transduction was carried out at a multiplicity of infection (MOI) of 20, with the addition of 6 μg/ml polybrene and the transduction enhancer surfactant Synperonic F108 (Sigma-Aldrich).

After fluorescence-activated cell sorting (FACS), the percentage of GFP-positive cells evaluated at flow cytometry was 93.15% for mock cells, 90.68% for NICD WT cells, and 89.89% for NICD-mutated cells ( Supplementary Figure S2A ). Western blotting and ddPCR were used to determine the NICD protein levels and NOTCH1 mutational status, respectively, of transduced PGA1 cells compared to those of untransduced PGA1 ( Supplementary Figures S2B, C ).

RNA was extracted using the RNeasy kit (Qiagen, Hilden, Germany), as per the manufacturer’s protocol, and complementary DNA (cDNA) synthesis was performed using an RT reagent kit (Takara Biotechnology Co., Ltd.). Quantitative real-time PCR (qPCR) was carried out using the 7300HT fast real-time PCR system (Applied Biosystems, Warrington, UK) and Power SYBR Green PCR master mix (Applied Biosystems, Warrington, UK). The relative mRNA expression levels were calculated using the 2^−ΔΔCq method and normalized to the internal control gene GAPDH. The sequences of primers are shown in Supplementary Table S3 .

Whole-cell lysates were extracted in a cold radioimmunoprecipitation assay (RIPA) lysis buffer containing a protease/phosphatase inhibitor cocktail (Sigma‐Aldrich). The protein concentration was determined by Bradford assay, and proteins were fractionated by sodium dodecyl sulfate–polyacrylamide gel electrophoresis (SDS-PAGE) and electro-transferred onto a nitrocellulose membrane (Millipore, MA, USA). Blots were blocked with 5% dried non-fat milk powder in Tris-buffered saline Tween 20 (TBST) and incubated overnight at 4°C against the primary antibodies listed in Supplementary Table S2 . After washing, membranes were incubated with horseradish peroxidase-conjugated secondary antibodies (Cell Signaling Technology, Beverly, MA) for 1 h and signals were obtained on a transilluminator (ChemiDoc™ MP imaging system, Bio-Rad Lab., Milan, Italy), and densitometric quantification was performed using the ImageLab software (Bio-Rad Lab.).

Cytosolic Ca²⁺ fluxes were determined by using the calcium-sensitive dye Fluo-4 Direct™ (Thermo Scientific), according to the manufacturer’s protocol. In brief, CLL primary cells were washed once to remove the medium. The cell pellet was then resuspended in Fluo-4 Direct™ calcium assay buffer at a density of approximately 2.5 × 10⁶ cells/ml. The plate was incubated at 37°C and 5% CO₂ for 60 min to allow the cells to settle. Subsequently, an equal volume of Fluo-4 Direct calcium reagent solution, containing 5 mM probenecid, was added to the cells, which were further incubated for an additional 45 min. Samples were acquired for 60 s to determine the basal levels of Ca²⁺. Then, either calcium ionophores (2 μM; A23187, Sigma-Aldrich), curcumin (15 μM), or DMSO (0.05%) was added to the cells, and data acquisition continued for an additional 420 s.

The acquisition was performed using a FACSCanto flow cytometer (BD Biosciences). The fluorescence intensity at 516 nm was monitored after excitation at 494 nm. The Ca²⁺ concentration was calculated, as described elsewhere (30), using the equation

where K_d represents the dissociation constant of calcium bound to the fluorochrome (as provided by the kit datasheet), R is the peak fluorescence observed in response to curcumin or the fluorescence at the basal level, R_min is the fluorescence measured at the end of acquisition (480 s), and R_max is the peak of fluorescence in response to ionophore.

Cell apoptosis was evaluated by flow cytometry after annexin V/propidium iodide (AnV/PI) double staining performed with a commercial kit (Immunotech, Beckman Coulter), according to the manufacturer’s instructions. Results were analyzed by FlowJo software version 10 (FlowJo, LLC, Ashland, OR, USA). For drug combination studies, CLL cells were incubated with different concentrations of curcumin (7.5, 15, and 30 µM) and venetoclax (1, 2, and 4 nM) alone and in combination. To calculate the combination index (CI) and the dose reduction index (DRI), data from AnV/PI assay were analyzed by using the CompuSyn software (ComboSyn Inc., Paramus, NJ, USA) (31). Based on the Chou–Talalay method, CompuSyn considers the complete shape of the growth inhibition curve. The CI was calculated using the formula CI = (D)1/(Dx)1 + (D)2/(Dx)2. (D)1 and (D)2 represent the doses of drug 1 and drug 2 in combination, and (Dx)1 and (Dx)2 represent the doses of drug 1 and drug 2 alone that produce the same effect level as the combination. A CI value less than 1 indicates synergism (greater than additive effect), a CI value equal to 1 suggests additivity (additive effect), and a CI value greater than 1 indicates antagonism (less than additive effect). The drug combination analysis of venetoclax and curcumin was conducted using a constant ratio of 1:7.5. In vivo experiments C57BL/6 (WT) and Eµ-TCL1 mice [provided by Prof. Paolo Ghia and generated by Prof. Carlo Croce (32)] were kept in specific pathogen-free conditions, and experiments were carried out in accordance with the protocols approved by the Italian Health Ministry (authorization no. 1155/2015-PR and no. 971/2020-PR). Eµ-TCL1 mice with 5% ± 3.25% of CD19⁺/CD5⁺ cells in PB received curcumin (50 mg/kg/day) or corn oil (vehicle) (Sigma-Aldrich), both administered by intraperitoneal injection for 2 months once daily. PB was analyzed by bi-monthly bleedings and at sacrifice. Bone marrow (BM), spleens, and livers were also processed and analyzed. BM and splenic cells were stained with anti-mouse CD19 and CD5 ( Supplementary Table S2 ) for cell sorting, performed using BD FACSAria™ III, considering 95% purity. All panel gates were drowned to exclude non-viable cells and debris.

Statistical analyses were performed using GraphPad Prism (GraphPad Software Inc., La Jolla, CA, USA). Statistical differences between mean values were evaluated using Student’s t-test for parametric data. Non-parametric data were analyzed by Wilcoxon (paired data) or Mann–Whitney (non-paired data) test. A one-way ANOVA test was used for the three groups’ comparison. Kaplan–Meier survival curves were compared using the Mantel–Cox rank sum test. Results were considered significant with a P-value< 0.05.

We first performed a bioinformatics analysis of NT1-M and NT1-WT CLL (21 vs. 31 samples, respectively) to identify DEGs, using three independent gene expression datasets (GSE75122, GSE137024, GSE92626). We found 620 DEGs in the comparison between NT1-M and NT1-WT CLL cells (470 upregulated; 150 downregulated; Supplementary Table S4 ). These DEGs were enriched for distinct Gene Ontology (GO) categories ( Supplementary Table S5 ), and among the increased set of biological processes, NOTCH1 mutation drove enrichment of the “immune system process” (P = 9.22 × 10⁻⁴²), “defense response” (P = 1.99 × 10⁻³¹), and “response to stress” (P = 9.62 × 10⁻²⁸) ( Figure 1A ). The downregulated categories evidenced enrichment for “cytoplasmic translation” (P = 4.53 × 10⁻²²), “translation” (P = 3.51 × 10⁻¹⁵), and the “peptide biosynthetic process” (P = 1.22 × 10⁻¹⁴). To identify altered canonical pathways, we performed an overrepresentation and topology of DEGs with the Reactome tool, which revealed a greater number of terms related to cellular responses to stimuli (circled in Figure 1B ), including “PERK regulates gene expression” and “ATF4 activates genes in response to ER stress” ( Supplementary Table S6 ). Genes involved in these pathways were further evaluated in a gene–gene interaction approach. Enrichment analysis revealed deregulation of the integrated stress response (ISR) (P = 1.8 × 10⁻⁷) and UPR (P = 2 × 10⁻⁵) genes in NT1-M CLL ( Figure 1C ).

Analysis of these genes in our CLL cells ( Supplementary Table S7 ) showed that ATF4, CHOP, and HSP90B1 mRNA expression was higher in NT1-M CLL compared to that in NT1-WT CLL (P = 0.008, P = 0.041, and P = 0.026, respectively) ( Figure 1D ).

In an attempt to better define the association of NT1 mutation with ER stress, we transduced the PGA1 CLL cell line with NICD WT or NICD-mutated fragment or an empty vector (mock). As shown in Supplementary Figure S2A , the percentages of PGA1 cells transduced with NICD WT or NICD-mutated fragment were 90.68 and 89.89, respectively. Western blot (WB) analysis of NICD showed that i) mock PGA1 cells express NICD WT levels similar to those of the untreated PGA1 control; ii) cells transduced with NICD WT overexpress NICD WT compared to mock cells; and iii) cells transduced with NICD-mutated fragment express NICD mutation, in addition to NICD WT ( Supplementary Figure S2B ). Using ddPCR multiplex assay, we confirmed NICD-mutated frequency in PGA1 transduced cells with NOCTH1 mutation ( Supplementary Figure S2C ). When we analyzed the effect of NICD transduction on ER stress markers, we found that PGA1 cells overexpressing NICD mutation showed increased expression of ATF4 and CHOP mRNA compared to both PGA1 mock (P = 0.021 and P = 0.0002, respectively) and PGA1 cells transduced with NICD WT (P = 0.06 and P = 0.0064, respectively), as shown in Figure 1E . Altogether, these data suggested that NOTCH1 mutation are associated with ISR/UPR/ER stress response markers in CLL.

Based on the above evidence that NOTCH1 mutation alters the expression of specific ER/UPR markers in CLL, we investigated whether NOTCH1 deregulation influenced CLL cell response to ER stress induction, by using curcumin, a natural ER stress inducer (33–36).

Curcumin-treated NT1-M cells expressed increased levels of PERK (P = 0.031), ATF4 (P = 0.031), BIP (P = 0.031), and HSP90B1 (P = 0.031) mRNAs compared to DMSO-treated cells, whereas GADD34 and HMOX1 mRNAs, essential in overcoming ER stress, were upregulated in NT1-WT cells (P = 0.031, both) ( Figure 2A ). After curcumin treatment, even CHOP mRNA was increased in NT1-M compared to NT1-WT cells ( Figure 2B ). Analysis of BiP/HSPA5 and CHOP proteins showed that curcumin significantly increased their levels in NT1-M cells compared to DMSO ( Figure 2C ). In NT1-WT, the levels of both proteins were not significantly changed, suggesting that in the absence of NOTCH1 mutation, the ER stress induced by curcumin is milder. We also examined the effect of curcumin on some homeostatic UPR markers. We analyzed i) the expression of IRE1α and ATF6 proteins, because either IRE1-mediated activation of XBP1 or activation of ATF6 induces the expression of chaperones and UPR components important for counteracting ER stress and restoring ER homeostasis (7, 12), and ii) the phosphorylation of eIF2α, a target of the PERK pathway, which leads to global inhibition of protein translation (8). Figure 2C shows that IRE1α levels were not affected by curcumin in NT1-WT CLL cells compared to controls, while they were reduced in NT1-M cells. eIF2α phosphorylation was increased by curcumin in CLL NT1-WT cells compared to that in controls, while it was reduced in NT1-M cells. Levels of ATF6, both full-length (FL) and the active cleaved (CL) fragment, were unaffected by curcumin either in NT1-WT or NT1-M CLL. Altogether, these results indicate that NOTCH1 mutation is associated with an exacerbated ER stress response after stimulation with an ER stress inducer, as also supported by experiments with thapsigargin ( Supplementary Figure S3 ).

Massive ER stress can trigger apoptosis via caspase 4 activation (13). When we analyzed the effect of curcumin on caspase 4 and its downstream caspase 3, we found that both caspases were cleaved after curcumin treatment, but cleavage started earlier and was higher in NT1-M than in NT1-WT CLL cells (N = 4; Figure 2D ).

Also, in NT1-M, but not in NT1-WT CLL cells, curcumin-induced apoptosis was preceded by a rapid and significant increase of [Ca²⁺]_i levels compared to DMSO (N = 6, P = 0.031) ( Figure 2E ). In keeping with this observation, it has been shown that [Ca²⁺]_i accumulation induced by high curcumin doses is accompanied by increased expression of UPR markers, including CHOP (35). Altogether, these results suggest that CLL harboring NOTCH1 mutation tends to be more sensitive to induced ER stress.

We then analyzed the effect of curcumin treatment for 24 h on the apoptosis of NT1-WT (N = 12), NT1-M^low (N = 13), and NT1-M^high (N = 11) CLL cells, using annexin V/PI assay. As shown in Figure 3A , curcumin treatment increased the apoptosis of NT1-WT (P = 0.0005), NT1-M^low (P = 0.0002), and NT1-M^high (P = 0.001) cells compared with that of the DMSO control. However, the percentage of increase in cell apoptosis induced by curcumin and normalized to that of the respective DMSO control was higher in both NT1-M^low (71.8% ± 53.1% vs. 35.1% ± 19.3%; P = 0.034) and NT1-M^high (63.3% ± 38.4% vs. 35.1% ± 19.3%; P = 0.05) than in NT1-WT CLL ( Figure 3A , right). These results suggest that CLL cells harboring NOTCH1 mutations appeared to be more prone to apoptosis than NT1-WT cells, regardless of mutational burden.

In line with flow cytometry data, WB analysis of the apoptotic marker PARP showed that its cleavage was increased by curcumin in NT1-M^low and NT1-M^high cells (P = 0.031, both) compared to that in the DMSO control (N = 6, for each group), whereas in NT1-WT cells (N = 6), curcumin did not significantly change PARP cleavage compared to that in DMSO (P = 0.218; Figure 3B ).

To investigate whether ER stress induced by curcumin in NT1-M CLL cells was associated with modulation of NOTCH1 signaling, we first measured NOTCH1 mRNA expression in primary CLL cells after 6 h of curcumin exposure. The NOTCH1 gene expression was significantly reduced in NT1-M cells (N = 7; P = 0.015) compared to that in the DMSO control but not significantly changed in NT1-WT cells (N = 7; P = 0.68) ( Figure 4A ). Accordingly, Figures 4B, C show that after curcumin treatment, there was a significant reduction in the mRNA expression of NOTCH1 downstream target genes HES1 and DTX1 in NT1-M (N = 7; P = 0.015, for both) compared to that in the DMSO control, respectively. Conversely, in NT1-WT CLL, curcumin induced a significant increase in DTX1 mRNA compared to that in DMSO (P = 0.031) but did not significantly affect HES1 levels. Next, we evaluated the effect of curcumin treatment for 24 h on the levels of the active NICD protein. As shown in Figure 4D , in NT1-M cells, NICD levels were significantly decreased by curcumin compared to that in the DMSO control (N = 9; P = 0.004) and were negatively correlated with the NOTCH1 mutation allelic ratio (P > 0.0001; Supplementary Figure S4 ). Conversely, in NT1-WT, curcumin induced a significant increase in NICD levels compared to that in the DMSO control (P = 0.0078). It has been shown that both NOTCH1 signaling and prolonged ER stress are associated with the expression of anti-apoptotic BCL2 family members, such as MCL1 (11, 36). WB analysis showed that in NT1-M CLL, curcumin reduced the levels of both MCL1 (P = 0.015) and BCL2 (P = 0.031) compared with those in DMSO, whereas in NT1-WT cells, it increased the levels of MCL1 (P = 0.015) without significantly influencing BCL2 expression ( Figure 4E ). In NT1-WT, the increase in MCL1 expression correlated positively with the increase in NICD levels (P = 0.021; Figure 4F ), reinforcing the relationship between NOTCH1 signaling and sustained MCL1 levels in CLL cells (37).

MCL1 expression and NOTCH activation also contribute to resistance to venetoclax therapy in CLL (38, 39). We tested the combined effect of curcumin (15 μM) and venetoclax (2 nM) on apoptosis, evaluated by AnV/PI assay, of NT1-M and NT1-WT CLL cells (N = 6, for both) after 24-h treatment. As shown in Figure 5A , in NT1-WT cells, the apoptotic effect induced by curcumin or venetoclax as single agents, compared to that in the DMSO control, was not significantly increased when the agents were used in combination. Conversely, in NT1-M cells, curcumin combined with venetoclax significantly increased the apoptotic effect induced by each single agent (P = 0.031 for both; Figure 5A ). The CI and isobologram of three NT1-M patients demonstrated a synergic effect for curcumin/venetoclax combination ( Figure 5B , Supplementary Figure S5 ). Based on the dose–effect curves of drugs alone and in combination, we calculated the DRI; we found that curcumin sensitized NT1-M CLL cells to the cytotoxic action of venetoclax and significantly decreased the efficacious dose of venetoclax.Curcumin elicited anti-leukemic effects associated with NOTCH1 inhibition in Eμ-TCL1 mice.

To better define the association between NOTCH1 signaling and ER stress, we used Eμ-TCL1 mice, a widely used mouse model for CLL, displaying an aberrant response to ER stress (14). B cells from the spleen and BM of homozygous Eμ-TCL1 mice expressed increased mRNA levels of ATF4 (P = 0.002, spleen; P = 0.04, BM) and CHOP (P = 0.002, spleen; P = 0.047, BM) ( Figure 6A ), as well as higher NOTCH1 downstream targets HES1 (N = 6; P< 0.01, spleen) and DTX1 (N = 6; P< 0.01, spleen; P< 0.01, BM) ( Figure 6B ) and NICD protein expression (P = 0.01, spleen; P = 0.05, BM; Figure 6C ), compared to B cells from C57BL6 WT and heterozygous Eμ-TCL1 mice ( Supplementary Figure S6 ).To examine whether curcumin was able to slow down the progression of CLL in Eμ-TCL1 mice, we treated mice with curcumin (50 mg/kg once every 2 days for 70 days) or vehicle ( Figure 6D ). While CD19⁺/CD5⁺ leukemic cells progressively expanded in the PB of vehicle-treated mice, curcumin treatment prevented their expansion, and this might be due to their higher apoptosis levels ( Supplementary Figure S7 ).

Figure 6E shows that CD19⁺/CD5⁺-infiltrating cells were significantly reduced in the spleen, liver, and BM (P = 0.05, P = 0.028, P = 0.028, respectively) of curcumin-treated mice. Splenic dimensions did not show a reduction when compared to those of the control group ( Supplementary Figure S8 ). Analysis of NOTCH1 activation in BM CD19⁺/CD5⁺ cells showed that NICD levels were reduced in curcumin-treated mice compared to those in vehicle-treated mice (N = 4, P = 0.028, Figure 6F ). The Kaplan–Meier survival curve revealed that vehicle-treated mice (N = 8, median survival = 142 days) succumbed significantly earlier than curcumin-treated mice (N = 8, m. s. = undefined, P = 0.04, Figure 6G ).