Targeting Nuclear NOTCH2 by Gliotoxin Recovers a Tumor-Suppressor NOTCH3 Activity in CLL

NOTCH signaling represents a promising therapeutic target in chronic lymphocytic leukemia (CLL). We compared the anti-neoplastic effects of the nuclear NOTCH2 inhibitor gliotoxin and the pan-NOTCH γ-secretase inhibitor RO4929097 in primary CLL cells with special emphasis on the individual roles of the different NOTCH receptors. Gliotoxin rapidly induced apoptosis in all CLL cases tested, whereas RO4929097 exerted a variable and delayed effect on CLL cell viability. Gliotoxin-induced apoptosis was associated with inhibition of the NOTCH2/FCER2 (CD23) axis together with concomitant upregulation of the NOTCH3/NR4A1 axis. In contrast, RO4929097 downregulated the NOTCH3/NR4A1 axis and counteracted the spontaneous and gliotoxin-induced apoptosis. On the cell surface, NOTCH3 and CD23 expression were mutually exclusive, suggesting that downregulation of NOTCH2 signaling is a prerequisite for NOTCH3 expression in CLL cells. ATAC-seq confirmed that gliotoxin targeted the canonical NOTCH signaling, as indicated by the loss of chromatin accessibility at the potential NOTCH/CSL site containing the gene regulatory elements. This was accompanied by a gain in accessibility at the NR4A1, NFκB, and ATF3 motifs close to the genes involved in B-cell activation, differentiation, and apoptosis. In summary, these data show that gliotoxin recovers a non-canonical tumor-suppressing NOTCH3 activity, indicating that nuclear NOTCH2 inhibitors might be beneficial compared to pan-NOTCH inhibitors in the treatment of CLL.

The conserved NOTCH gene family (NOTCH1-4) encodes transmembrane receptors that regulate a wide variety of differentiation processes by modulating binary cell fate decisions in response to external signals [19][20][21][22]. Canonical NOTCH signaling is induced by ligand binding, the intracellular domain of NOTCH (N IC ) is released by a series of proteolytic events involving γ-secretase followed by translocation to the nucleus, where it acts as context and cell type specific transcription factor on CSL (for CBF1, Suppressor of Hairless, and LAG-1)-responsive genes like FCER2 (CD23) in CLL cells [7,18,[20][21][22]. However, non-canonical NOTCH signaling also exists and involves the activation of NFκB [23]. In the murine system, Notch2 is implicated in the development of marginal zone (MZ) B2 B-cells and of Cd5+ (B-1a) B-lymphocytes [24], and is indispensable for CLL initiation in Cd5+ (B-1a) B-cells [25].
Deregulation of NOTCH signaling is observed in an increasing number of human neoplasms, where the individual NOTCH receptors act either as oncogenes or as tumor suppressors, depending on the cellular context and microenvironment [20,26,27]. Therefore, targeting oncogenic NOTCH, for example with γ-secretase inhibitors (GSI), represents a promising therapeutic strategy in the treatment of NOTCH-associated tumors/leukemias [27][28][29][30][31]. In a first attempt to address this issue, we found that the majority of CLL cases express GSI-resistant NOTCH2/CSL transcription factor complexes and did not respond to the selective GSI DAPT [18]. In contrast, targeting nuclear NOTCH2 with the Aspergillum-derived NOTCH2/CSL transactivation inhibitor, gliotoxin efficiently induced apoptosis in CLL cells by a mechanism involving the induction of the NOTCH3 and the NR4A1 gene on the mRNA level [32]. However, the global effect of gliotoxin on the complex and interconnected signal transduction pathways and the role of NOTCH3 in CLL cells remains to be determined.
In the current study, we extended our prior work and compared the anti-neoplastic effects of gliotoxin and the GSI RO4929097 [29,31,33] in a reasonable cohort of well-characterized CLL cases. Here we show that the inhibition of NOTCH2 signaling by gliotoxin is associated with the recovery of a potentially non-canonical tumor suppressing NOTCH3 activity in CLL cells. Furthermore, assays for transposase-accessible chromatin with high-throughput sequencing (ATAC-seq) revealed that gliotoxin treatment is associated with prominent changes in the epigenetic landscape in CLL cells.

Gene Silencing by RNA-Interference
The siRNA duplexes (siRNAs) for NOTCH3 (ON-TARGETplusTM) and the controls (RISC-free Co-siRNA, and siGLO red transfection indicator) were obtained from Dharmacon (Lafayette, CO, USA). Transfection of siRNAs into the CLL cells was performed by using the lipid reagent siLentFect TM from Bio-Rad Laboratories (Hercules, CA, USA). The transfection efficiency was determined by FACS and varied from 70 to 90%.

ATAC-Seq
Accessible-chromatin mapping on the CLL cells was performed using the ATAC-seq method with minor modifications together with the ATAC-seq processing pipeline, as described previously [34,35]. Principle component analysis (PCA) was performed on the quantile-normalized log-transformed values of chromatin accessibility across all the accessible sites discovered in all samples, and DESeq2 was used to detect the differential regions between treatment timepoints across patients [36]. Regions with an FDR-adjusted p-value smaller than 0.05 were selected and clustered using the Euclidean distance and complete linkage, from which two clusters representing the earliest branching point were extracted. HOMER [37] was used for de novo motif finding on the region clusters and LOLA [38] for enrichment in previously existing location-based datasets. Genes assigned to the two clusters of regions were enriched using the Enrichr tool [39].

Dose and Time-Dependent Effects of Gliotoxin and GSI on CLL Cell Viability
Eighteen CLL cases (Table 1) were subjected to MTT assays and the dose-dependent effect of gliotoxin, RO4929097, and DAPT (0.01 to 10µM) on cell viability was determined after 3 and 7 days of incubation.
After 3 days, gliotoxin efficiently decreased CLL cell viability (IC50 between 0.1 µM and 1 µM), whereas RO2949097 and DAPT were only partly effective in a subset of patient samples, including those cases characterized by the NOTCH1∆CT mutation ( Figure 1A-C). The mean inhibition of CLL cell viability (±SD) in the NOTCH1 wild type versus NOTCH1∆CT mutated CLL cases at an inhibitor concentration of 0.5 µM after 3 days was as follows: 84% (±23%) versus 87% (±16%) for gliotoxin, 14% (±13%) versus 36% (±14%) for RO4929097, and 10% (±14%) versus 33% (±16%) for DAPT, respectively. The dose-and time-dependent effect of gliotoxin, RO4929097, and DAPT on CLL cell viability in vitro. Eighteen CLL cases, including 4 cases with the recurrent NOTCH1ΔCT mutation (black lines), 6 cases expressing GSI-R nuclear NOTCH2 (pink lines), and 5 cases expressing GSI-S nuclear NOTCH2 (blue lines), were subjected to MTT assays and the relative inhibition of cell viability in response to the indicated drug concentrations was determined after 3 (A-C) and 7 days (D-F), respectively. The percent inhibition relative to the controls was calculated from the mean OD (optical density) values from CLL samples cultured in triplicates. The IC50 of the individual compounds is indicated. Figure 1. The dose-and time-dependent effect of gliotoxin, RO4929097, and DAPT on CLL cell viability in vitro. Eighteen CLL cases, including 4 cases with the recurrent NOTCH1∆CT mutation (black lines), 6 cases expressing GSI-R nuclear NOTCH2 (pink lines), and 5 cases expressing GSI-S nuclear NOTCH2 (blue lines), were subjected to MTT assays and the relative inhibition of cell viability in response to the indicated drug concentrations was determined after 3 (A-C) and 7 days (D-F), respectively. The percent inhibition relative to the controls was calculated from the mean OD (optical density) values from CLL samples cultured in triplicates. The IC 50 of the individual compounds is indicated. After 7 days, the response to RO4929097 was more prominent and 11 out of the 18 treated CLL cases (61%) reached IC 50 levels at a drug concentration of 0.5 µM ( Figure 1E) and, thus, were phenotypically considered as GSI sensitive. The NOTCH1∆CT mutated CLL cases and/or CLL cases expressing the GSI-sensitive (GSI-S) nuclear NOTCH2 clustered in this GSI-sensitive group. In contrast, CLL cases expressing GSI-resistant (GSI-R) nuclear NOTCH2 and with the wild type (wt) NOTCH1 status clustered in the GSI-resistant group. Maximal responses were already achieved at 0.1 µM RO4929097 and increasing amounts did not significantly enhance its effect on the CLL cells. The drop in CLL cell viability at higher drug concentrations (10 µM) after this plateau phase might be attributed to off-target effects ( Figure 1B,E) [40]. The mean inhibition of CLL cell viability (±SD) in the NOTCH1wT versus NOTCH1∆CT mutated CLL cases at an inhibitor concentration of 0.5 µM after 7 days was as follows: 95% (±6%) versus 92% (±5%) for gliotoxin, 40% (±20%) versus 54% (±7%) for RO4929097, and 22% (±17%) versus 38% (±12%) for DAPT, respectively.

GSI Inhibited Spontaneous Apoptosis in Early Clinical Stage-Derived CLL Samples Expressing GSI Resistant Nuclear NOTCH2
The observed delayed effect of GSI on CLL cell viability in the MTT assays suggests that GSI primarily affect the metabolic activity rather than directly inducing apoptosis. Therefore, we measured the percentage of apoptotic CLL cells (n = 16) after exposure to equal doses (0.5 µM) of gliotoxin, RO4929097, and DAPT by AnnexinV/PI staining.

The Inhibition of Spontaneous Apoptosis by RO4929097 is Associated with Inhibition of Recovered NOTCH3 mRNA Expression in CLL Cells
We next confirmed in a time kinetic RT-PCR experiment in two additional representative CLL

The Inhibition of Spontaneous Apoptosis by RO4929097 is Associated with Inhibition of Recovered NOTCH3 mRNA Expression in CLL Cells
We next confirmed in a time kinetic RT-PCR experiment in two additional representative CLL samples (CLL21, 22) our previous observation that gliotoxin inhibited NOTCH2, to a lesser extent NOTCH1, and induced NOTCH3 mRNA expression within 4 h of incubation ( Figure 3A) [32]. Interestingly, RO4929097 counteracted gliotoxin-induced NOTCH3 transcription after one day ( Figure 3A). This suggests that RO4929097 interrupted a positive feedback loop of NOTCH3 mRNA expression. NOTCH receptors regulate context and cell type specific their own expression and each other in positive and negative feedback loops [19]. Moreover, RO4929097 decreased (58 versus 64%) or increased (86 versus 79%) the effect of gliotoxin on the percentage of apoptotic cells, depending on the GSI sensitivity of nuclear NOTCH2 ( Figure 3A). Therefore, we asked whether NOTCH3 might also account for the GSI-mediated inhibition of spontaneous apoptosis in early clinical stage-derived CLL long-term suspension cultures.
NOTCH3 mRNA was not detectable in frozen samples from freshly isolated CLL cells of our initial drug screening cohort ( Figure 3C). After 7 days, however, the anti-apoptotic effect of RO4929097 in two representative Rai/Binet I/A patient samples expressing GSI-R nuclear NOTCH2 (CLL8 and CLL9, Figure 3B) was clearly associated with inhibition of spontaneously recovered NOTCH3 mRNA expression, together with unchanged or even enhanced NOTCH2 mRNA and NOTCH2/CSL DNA-complex expression ( Figure 3C, left panel). with a decrease in NOTCH2 mRNA and NOTCH2/CSL DNA-complex expression without any detectable NOTCH3 gene activity after 7 days in culture ( Figure 3C, right panel). In accordance with published data [9,16], the NOTCH1 mRNA was GSI sensitive in all cases and was more expressed in the Rai/Binet IV/C-derived CLL cells ( Figure 3C). Together, this combined approach suggests that recovery of NOTCH3 mRNA expression is involved in the inhibition of spontaneous apoptosis by GSI in CLL long-term suspension cultures and may be associated with GSI resistance of NOTCH2 and early stage derived CLL samples.

Induction of Surface NOTCH3 Expression by Gliotoxin is Associated with Downregulation of CD23 and Increased Apoptosis of CLL Cells
We next analyzed NOTCH3 and FCER2 (CD23) expression in CLL cells (n = 4) in relation to spontaneous as well as gliotoxin induced apoptosis on the mRNA and protein level by RT-PCR and FACS (Figure 4; see Supplemental Figure 1 for a detailed FACS analysis of CLL24 cells). NOTCH3 was almost undetectable on the mRNA ( Figure 4A) and on the cell surface protein level ( Figure 4B) in freshly isolated CLL cells. After 4 days in culture, we found an increase of surface NOTCH3 expression together with a decrease in surface CD23 on CLL23 and CLL24 cells ( Figure 4B). On the mRNA level, NOTCH3 expression was below the detection limit in the 4 days control. The loss of CD23 expression seemed to be a prerequisite for NOTCH3 expression since NOTCH3 was primarily detected on CD23-negative CLL cells. In contrast, the induction of apoptosis by RO4929097 in two representative Rai/Binet IV/C patient samples expressing GSI-S nuclear NOTCH2 (CLL10 and CLL16, Figure 3B) was associated with a decrease in NOTCH2 mRNA and NOTCH2/CSL DNA-complex expression without any detectable NOTCH3 gene activity after 7 days in culture ( Figure 3C, right panel). In accordance with published data [9,16], the NOTCH1 mRNA was GSI sensitive in all cases and was more expressed in the Rai/Binet IV/C-derived CLL cells ( Figure 3C).
Together, this combined approach suggests that recovery of NOTCH3 mRNA expression is involved in the inhibition of spontaneous apoptosis by GSI in CLL long-term suspension cultures and may be associated with GSI resistance of NOTCH2 and early stage derived CLL samples.

Induction of Surface NOTCH3 Expression by Gliotoxin is Associated with Downregulation of CD23 and Increased Apoptosis of CLL Cells
We next analyzed NOTCH3 and FCER2 (CD23) expression in CLL cells (n = 4) in relation to spontaneous as well as gliotoxin induced apoptosis on the mRNA and protein level by RT-PCR and FACS (Figure 4; see Supplemental Figure S1 for a detailed FACS analysis of CLL24 cells). NOTCH3 was almost undetectable on the mRNA ( Figure 4A) and on the cell surface protein level ( Figure 4B) in freshly isolated CLL cells. After 4 days in culture, we found an increase of surface NOTCH3 expression together with a decrease in surface CD23 on CLL23 and CLL24 cells ( Figure 4B). On the mRNA level, NOTCH3 expression was below the detection limit in the 4 days control. The loss of CD23 expression seemed to be a prerequisite for NOTCH3 expression since NOTCH3 was primarily detected on CD23-negative CLL cells. which might be attributed to the loss of the inhibitory effect of these drugs on CD23 over time ( Figure  4A,B). Gliotoxin induced the NOTCH3 gene ( Figure 4A), enhanced NOTCH3 surface expression, and upregulated apoptosis in a dose-dependent manner in all cases ( Figure 4B). The remaining living CLL cells were enriched for CD23-positive and NOTCH3-negative cells ( Figures 4B,C and S1), confirming that CD23 expression is associated with CLL cell viability whereas NOTCH3 expression is associated with CLL cell apoptosis.
In summary, we found a direct correlation between the percentage of surface NOTCH3-positive and apoptotic CLL cells and an indirect correlation of these two parameters with the percentage of CD23-positive and living CLL lymphocytes ( Figure 4D).  In contrast, we detected a low expression of CD23 on freshly isolated cells from Duvelisib (CLL25) or Ibrutinib (CLL26) treated CLL patients ( Figure 4B). However, FCER2 (CD23) expression spontaneously recovered on the mRNA and protein level in these 2 samples after 4 days in culture which might be attributed to the loss of the inhibitory effect of these drugs on CD23 over time ( Figure 4A,B).

Targeting NOTCH3 Signaling Decreased NR4A1 mRNA Expression and Counteracted Gliotoxin Induced Apoptosis in CLL Cells
Gliotoxin induced the NOTCH3 gene ( Figure 4A), enhanced NOTCH3 surface expression, and upregulated apoptosis in a dose-dependent manner in all cases ( Figure 4B). The remaining living CLL cells were enriched for CD23-positive and NOTCH3-negative cells ( Figure 4B,C and Figure S1), confirming that CD23 expression is associated with CLL cell viability whereas NOTCH3 expression is associated with CLL cell apoptosis.
In summary, we found a direct correlation between the percentage of surface NOTCH3-positive and apoptotic CLL cells and an indirect correlation of these two parameters with the percentage of CD23-positive and living CLL lymphocytes ( Figure 4D).

Targeting NOTCH3 Signaling Decreased NR4A1 mRNA Expression and Counteracted Gliotoxin Induced Apoptosis in CLL Cells
We hypothesized that NOTCH2 and NOTCH3 have opposite roles in the binary cell fate decision between positive and negative selection of the activated CLL cells. Therefore, we investigated the effects of targeting NOTCH3 by RO4929097, or more specifically, by siRNA in PMA-stimulated CLL cells [32]. In this model, CLL cells form tight clusters and express the CLL proliferation center marker MYC ( Figure 5B), resembling the situation found in lymphoid tissues [41,42]. To avoid background effects on NOTCH2 signaling, we selected five CLL cases expressing GSI-R NOTCH2 (CLL7, 8,9,13,21).
A time kinetic confirmed that CD23 was expressed on almost all representative CLL9 cells after one day of PMA stimulation ( Figure 5A) [18]. Interestingly, prolonged stimulation with PMA for 3 days led to the downregulation of CD23, upregulation of NOTCH3, and an increase in the percentage of apoptotic CLL9 cells. This effect was clearly enhanced by gliotoxin treatment ( Figure 5A).
NOTCH3 gene silencing by siRNA decreased the apoptotic effect of the gliotoxin, leading to a 3.5-fold increase in (Ax-/PI-) living CLL9 cells with increased surface CD23 expression after 3 days in culture ( Figure 5D). Corresponding RT-PCR analysis confirmed that NOTCH3 gene silencing downregulated the NOTCH3/NR4A1 axis and upregulated the NOTCH2/FCER2 (CD23) axis ( Figure 5E), resembling the effect of GSI treatment ( Figure 5B, right panel). Interestingly, gliotoxin-mediated upregulation of the NR4A1 gene was completely blocked by an NFκB activation inhibitor ( Figure 5F), suggesting that NOTCH3 regulates NR4A1 transcription via non-canonical NOTCH3/NFκB signaling [23,46,47]. This would explain the lack of DNA-bound NOTCH3/CSL complexes in EMSA. All other control experiments show similar trends.
Collectively, these data strongly suggest that inhibition of the anti-apoptotic canonical NOTCH2/CSL signaling (NOTCH2/FCER2 axis) by gliotoxin recovers a GSI sensitive pro-apoptotic non-canonical NOTCH3 function which may involve NFκB dependent NR4A1 expression (NOTCH3/NR4A1 axis) in CLL cells. A hypothetical model summarizing the proposed counteracting roles of NOTCH2 and NOTCH3 in CLL cells is given in Figure 5G.

Gliotoxin Modulates Chromatin Accessibility at Gene Regulatory Elements Containing Potential NOTCH/CSL and NR4A1 Binding Sites
We have recently shown that genome-wide mapping of gene-regulatory elements using the transposase-accessible chromatin (ATAC-seq) assay is a useful tool to investigate gene regulation in CLL cells [35]. Therefore, we analyzed gliotoxin-induced chromatin changes in CLL cells after 3 days of incubation (n = 7, Figure 6A), where we identified 62,760 unique chromatin accessible regions ( Figure S2A). These sites represent mainly enhancers and promoters ( Figure S2B, a representative locus spanning the CXCR4 gene region is shown in Figure 6B). Unsupervised principal component analysis (PCA) confirmed that IGHV mutation status is the major source of heterogeneity in chromatin accessibility in CLL cells, as described previously ( Figure 6C, left panel) [35]. However,

Gliotoxin Modulates Chromatin Accessibility at Gene Regulatory Elements Containing Potential NOTCH/CSL and NR4A1 Binding Sites
We have recently shown that genome-wide mapping of gene-regulatory elements using the transposase-accessible chromatin (ATAC-seq) assay is a useful tool to investigate gene regulation in CLL cells [35]. Therefore, we analyzed gliotoxin-induced chromatin changes in CLL cells after 3 days of incubation (n = 7, Figure 6A), where we identified 62,760 unique chromatin accessible regions ( Figure S2A). These sites represent mainly enhancers and promoters ( Figure S2B, a representative locus spanning the CXCR4 gene region is shown in Figure 6B). Unsupervised principal component analysis (PCA) confirmed that IGHV mutation status is the major source of heterogeneity in chromatin accessibility in CLL cells, as described previously ( Figure 6C, left panel) [35]. However, Principle Components 1 and 4 clearly revealed dose-dependent changes in chromatin accessibility in response to gliotoxin treatment ( Figure 6C, right panel).
Clustering of significantly changing regions ( Figure 7A) segregated the sites into those that lose accessibility (Region Cluster 1, marked in blue) and those that gain accessibility (Region Cluster 2, marked in orange) in response to gliotoxin treatment (representative genomic loci of each cluster are shown in Figure 7B). By overlapping the differential ATAC-seq sites with publicly available transcription factor binding experiments [38], we observed that Region Cluster 1 largely overlaps with the functional NOTCH/CSL binding sites in T-cell acute leukemia cells ( Figure S3A). On the other hand, Region Cluster 2 largely overlaps with the functional NFκB binding sites in lymphoblastoid cell lines ( Figure S3B).
To get unbiased insights into the potential regulators of differentially accessible regions we carried out de novo motif analysis to discover the enriched DNA binding elements. Differential accessible gene-regulatory elements in Region Cluster 1 were found to be enriched for the promoter-associated motif GFY (general factor Y) and the recognition site for ZNF143, both of which contain the NOTCH/CSL consensus binding site TGGGAA [48]. In Region Cluster 2, we found an enrichment of consensus sites for the transcription factors ATF3, PU.1, NFκB, FLI1, NR4A1 (nur77), and RUNX ( Figure 7C). Clustering of significantly changing regions ( Figure 7A) segregated the sites into those that lose accessibility (Region Cluster 1, marked in blue) and those that gain accessibility (Region Cluster 2, marked in orange) in response to gliotoxin treatment (representative genomic loci of each cluster are shown in Figure 7B). By overlapping the differential ATAC-seq sites with publicly available transcription factor binding experiments [38], we observed that Region Cluster 1 largely overlaps with the functional NOTCH/CSL binding sites in T-cell acute leukemia cells ( Figure S3A). On the other hand, Region Cluster 2 largely overlaps with the functional NFκB binding sites in lymphoblastoid cell lines ( Figure S3B).
To get unbiased insights into the potential regulators of differentially accessible regions we carried out de novo motif analysis to discover the enriched DNA binding elements. Differential accessible gene-regulatory elements in Region Cluster 1 were found to be enriched for the promoterassociated motif GFY (general factor Y) and the recognition site for ZNF143, both of which contain the NOTCH/CSL consensus binding site TGGGAA [48]. In Region Cluster 2, we found an enrichment of consensus sites for the transcription factors ATF3, PU.1, NFκB, FLI1, NR4A1 (nur77), and RUNX ( Figure 7C).

Discussion
In the last decade, much effort has been spent into the development of GSIs as tools for therapy for NOTCH-associated human neoplasias [29][30][31]. However, four NOTCH family members are present in mammalian cells, where the individual receptors may have opposite functions concerning their role as oncogenes or tumor suppressors in a context-and microenvironment-dependent manner [26,30,31]. Moreover, transformed cells may express truncated NOTCH forms that do not depend on γ-secretase for processing and function [49,50]. A search for an alternative to GSI revealed that the Aspergillum-derived secondary metabolite gliotoxin is a potent inhibitor of DNA-bound NOTCH2/CSL transcription factor complexes, and efficiently induced apoptosis in CLL lymphocytes and in NOTCH2-associated solid tumor cell lines-independent of their sensitivity to GSIs [32,51] In this study, we show that gliotoxin as a nuclear NOTCH2 inhibitor may have an additional therapeutic advantage over GSI as pan-NOTCH inhibitors in CLL. We were able to demonstrate that the GSI RO4929097 targets an unexpected non-canonical tumor-suppressing NOTCH3 activity, which is involved in spontaneous as well as gliotoxin-induced apoptosis in CLL cells.
On the CLL cell surface, upregulation of NOTCH3 was associated with downregulation of CD23, suggesting that downregulation of the NOTCH2/FCER2 (CD23) axis, either spontaneously due to the lack of appropriate activation stimuli in vitro [18], or by gliotoxin treatment [32], is a prerequisite for NOTCH3 expression in CLL cells. The strong association between surface NOTCH3 expression and apoptosis suggests that NOTCH3 signaling is involved in the execution phase of CLL cell apoptosis by prolonging the expression of the newly identified NOTCH3 target gene NR4A1, a multi-functional tumor-suppressor gene implicated in the regulation of B-cell tolerance to self-antigens [52][53][54].
ATAC-seq confirmed that gliotoxin targets canonical NOTCH signaling as indicated by reduced accessibility at potential NOTCH/CSL consensus sites (TGGGAA) [48]. This includes the promoter-associated motif GFY, and the consensus site for ZNF143, a transcription factor known to modulate NOTCH target gene expression in competition with CSL [55]. In contrast, gliotoxin increased the chromatin accessibility at the potential ATF3, PU.1, NFκB, FLI1, RUNX, and NR4A1 binding sites in the vicinity of genes involved in B-cell activation, differentiation, and apoptosis. This might reflect, at least in part, the loss of a NOTCH2-mediated differentiation arrest and the induction of non-canonical NOTCH3/NFκB signaling [23]. It has been shown that NOTCH3 activates NFκB, a positive regulator of the NR4A1 gene [46], through an IKKα-dependent alternative pathway [47]. This would not only explain the gain in chromatin accessibility at NFκB and NR4A1 consensus sites and the lack of DNA-bound NOTCH3/CSL complexes in response to gliotoxin treatment, but also the NOTCH3-dependent NR4A1 gene activity in CLL cells. In line with this assumption, we found that inhibition of NFκB counteracted gliotoxin-mediated upregulation of NR4A1 mRNA in CLL lymphocytes. However, this important issue needs further mechanistic exploration.
Under physiological conditions, NOTCH2 signaling might be induced by ligand-expressing surrounding cells in order to protect the proliferative/regenerative reservoir of CD5+ B-cells from NR4A1-mediated activation-induced cell death (AICD) [32,52,60]. This scenario might take place in the marginal zone (MZ) of the spleen [61], where DLL1-expressing bystander cells have been identified [62]. In contrast, apoptotic NOTCH3 signaling might counteract the uncontrolled expansion of CD5+ B-cells in the periphery. In line with this hypothesis, we found that NOTCH3 inhibition by RO4929097, or more specifically, by siRNA, downregulated the NOTCH3/NR4A1 axis, enhanced the NOTCH2/FCER2 (CD23) axis, and counteracted apoptosis in CLL cells. The NOTCH3 gene is frequently epigenetically silenced in B-cell acute lymphocytic leukemia (B-ALL) cells, pointing to a broader tumor-suppressor role of NOTCH3 in B-cells [63]. Moreover, the CLL downregulated/deleted MicroRNA-16 [64] has been shown to exert its pro-apoptotic function by NOTCH2 inhibition in pre-eclampsia, where an inverse correlation between NOTCH2 and NOTCH3 expression also has been found [59].
In terms of CLL biology, constitutive active NOTCH2 might dominantly suppress apoptotic NOTCH3 signaling, thereby enabling the (self-) antigen-driven progredient expansion of the malignant clone. The affinity and avidity of the corresponding (self-) antigens might be the basis for the NOTCH2 dominance seen in CLL cells. A short time exposure to B-cell activation mimetic PMA favors the NOTCH2/FCER2 (CD23) axis, whereas long-term stimulation with PMA shifts the NOTCH2/FCER2 (CD23) axis to the NOTCH3/NR4A1 axis in CLL cells. One likely mediator of this bi-phasic PMA effect might be the activation and subsequent downregulation of the B-cell receptor-associated protein kinase C-delta (PKC-δ) [65][66][67], a positive regulator of nuclear NOTCH2 activity [10,18,27,68]. Therefore, compounds that interfere with B-cell activation might affect the expression of both NOTCH receptors in CLL cells. This would explain why CLL cells pre-treated with Duvelisib (a dual PI3K-δ/γ inhibitor) [69] or Ibrutinib (a Bruton's tyrosine kinase inhibitor) [2,44] express lower amounts of CD23 and NOTCH3 and are less sensitive to apoptosis.
The additional effect of NOTCH1 signaling in this scenario is less clear. NOTCH1 is not detectable in nuclear NOTCH/CSL transcription factor complexes in CLL cells [7,8,18,32]. However, NOTCH1 is frequently mutated and/or overexpressed in advanced stage CLL cells, where it has a CLL-driving role by regulating MYC expression [8,9,16,17,[43][44][45]. In this context, NOTCH1 may indirectly account for the relative GSI sensitivity of NOTCH2, keeping in mind that active NOTCH1 is a positive regulator of the NOTCH2 gene in CLL cells ( Figure 5G) [12].

Conclusions
In summary, we show that the nuclear NOTCH2 inhibitor gliotoxin has global effects on the NOTCH signaling network in CLL cells, including the recovery of a newly identified non-canonical tumor suppressing NOTCH3 activity. This proof-of-concept may be the basis for the design of innovative therapies aimed at specifically targeting oncogenic NOTCH signaling in CLL cells.