Search

MALT1 inhibitors are promising therapeutic agents for B-cell lymphomas that are dependent on constitutive or aberrant signaling pathways. However, a potential limitation for signal transduction–targeted therapies is the occurrence of feedback mechanisms that enable escape from the full impact of such drugs. Here, we used a functional genomics screen in activated B-cell–like (ABC) diffuse large B-cell lymphoma (DLBCL) cells treated with a small molecule irreversible inhibitor of MALT1 to identify genes that might confer resistance or enhance the activity of MALT1 inhibition (MALT1i). We find that loss of B-cell receptor (BCR)- and phosphatidylinositol 3-kinase (PI3K)-activating proteins enhanced sensitivity, whereas loss of negative regulators of these pathways (eg, TRAF2, TNFAIP3) promoted resistance. These findings were validated by knockdown of individual genes and a combinatorial drug screen focused on BCR and PI3K pathway–targeting drugs. Among these, the most potent combinatorial effect was observed with PI3Kδ inhibitors against ABC-DLBCLs in vitro and in vivo, but that led to an adaptive increase in phosphorylated S6 and eventual disease progression. Along these lines, MALT1i promoted increased MTORC1 activity and phosphorylation of S6K1-T389 and S6-S235/6, an effect that was only partially blocked by PI3Kδ inhibition in vitro and in vivo. In contrast, simultaneous inhibition of MALT1 and MTORC1 prevented S6 phosphorylation, yielded potent activity against DLBCL cell lines and primary patient specimens, and resulted in more profound tumor regression and significantly improved survival of ABC-DLBCLs in vivo compared with PI3K inhibitors. These findings provide a basis for maximal therapeutic impact of MALT1 inhibitors in the clinic, by disrupting feedback mechanisms that might otherwise limit their efficacy.

MALT1 inhibition is a promising strategy against B-cell receptor (BCR)-dependent lymphomas including ABC DLBCL, CLL and MCL. MALT1 is downstream of the most frequently mutated genes in the BCR and Toll-like receptor (TLR) pathways. MALT1 inhibitors are active in Ibrutinib-resistant BTK and PLCγ2 mutant CLL or CARD11 mutant ABC DLBCL. Therefore, MALT1 inhibitors, which recently began first in man clinical testing, have the potential to cover a larger patient population than drugs against more upstream targets in the BCR pathway including BTK inhibitors. However, activity of inhibitors targeting signaling mediators can be limited by feedback mechanisms counteracting and/or bypassing the need for a specific pathway. MALT1 is central to NF-κB activation downstream of the BCR. Therefore, activation of alternative pathways leading to full or partial activation of the BCR program or other pro-survival pathways might enable cell survival and set off MALT1i resistance. In order to map the landscape of resistance/sensitivity to MALT1 inhibition in ABC DLBCL and assist design of combinatorial regimens, we carried out a loss-of-function screen in HBL-1 (MALT1i sensitive) with or without MI-2 (small molecule irreversible inhibitor of MALT1) to identify genes capable of modulating response to MALT1 inhibition. Our analyses showed that loss of BCR and PI3K activators enhanced sensitivity, while loss of negative regulators of these pathways promoted MALT1i resistance. These findings were validated by knockdown of individual genes with two independent hairpins against activators CD79B, CARD11, BTK or the negative regulator TNFAIP3. Next, we carried out a combinatorial drug screen anchored in MALT1 inhibition by MI-2 or C3 (irreversible substrate-mimetic MALT1 inhibitor) and focused on inhibitors against signaling hubs in the BCR/PI3K pathways in 4 MALT1i sensitive cell lines. This combinatorial screen confirmed that concurrent inhibition of MALT1 and other BCR and PI3K pathways' proteins is additive (0.9>CI (combination index) In depth analysis of proliferation and cell death by CFSE dilution and Annexin V staining revealed that both MALT1/PI3K-i and MALT1/MTORC1-i combinations significantly enhanced growth inhibition and apoptosis in TMD8 and HBL-1 compared to individual agents. Results with MI-2 and C3 were comparable. Short exposure to MI-2 or C3 increased MTORC1 activity as assessed by S6K-Thr389 and S6-Ser235/6 phosphorylation in TMD8 and HBL-1 indicating that MALT1 protease activity modulates MTORC1 activation. MTOR activation is tumorigenic and can mediate chemotherapy resistance. Increased p-S6 following MALT1 inhibition (FC=1.4-2) was blocked by Idelalisib (PI3Ki). However, only Rapamycin, an MTORC1 inhibitor, reduced p-S6 levels relative to vehicle, FC=-5 alone or in combination. In vivo, MALT1/PI3K-i (MI-2/Idelalisib) significantly delayed tumor progression compared to single treatments (p MALT1/MTORC1-i regimens were also highly synergistic in 2 patient-derived xenografted (PDX) ABC DLBCL ex vivo. PDXs were cultured in gelatin/silicate nanoparticle hydrogel 3-D organoids and co-cultured with CD40L expressing cells. Compound pairs were assayed for synergy using 4x4 matrices and growth inhibition evaluated by flow cytometry. Synergy ZIP δ-score ranged 7-14 in 2 specimens for 2 MALT1i. Combined, these results suggest that: 1) MTORC1 activation constitutes a survival feedback mechanism activated after MALT1i treatment that might be leveraged by tumoral cells to evade MALT1 inhibition and, 2) that simultaneous targeting of MTORC1 could improve response and prevent resistance to MALT1 inhibitors. Disclosures Melnick: Constellation: Consultancy; Janssen: Research Funding; Epizyme: Consultancy.

Recent studies have identified small molecule inhibitors of the paracaspase activity of MALT1, a protease and scaffolding protein involved in the B-cell receptor (BCR) signaling pathway, that are effective killing lymphomas in vitro and in vivo in xenograft models of Activated B-cell like Diffuse Large B-cell Lymphoma (ABC-DLBCL). ABC-DLBCL is characterized by constitutive NF-κB activity. This activation has been attributed to mutations in various protein components of the B-cell receptor (BCR) as well as Toll-like receptor (TLR) pathways. However, not all ABC-DLBCL cell lines and primary patient samples were equally sensitive to MALT1 inhibitors in vitro. In order to discover genetic modifiers of response to MALT1 inhibition we used an shRNA library screening approach. MALT1 inhibition sensitive cell line HBL-1 was infected with DECIPHER barcoded shRNA library Module 1 and cells were treated with vehicle or 300 nM of MALT1 inhibitor MI-2 for 22 days. At this time cells were harvested and genomic DNA extracted. PCR was used to amplify barcodes and gel purified bands were extracted and sequenced. Cellecta's Deconvoluter software was used to quantify the number of reads per shRNA, reads were normalized to total number of reads and fold change between vehicle and MI-2 treated cells was calculated. Among the top positively and negatively enriched hairpins, we found a significant number of genes involved in the BCR pathway including: positively regulated shRNAs against TNFAIP3 and FOXO1 and negatively regulated hairpins against BTK, CD79B and PI3K genes PIK3C2A and PIK3C2D. Interestingly, TNFAIP3 and FOXO1 are negative regulators of the BCR pathway while BTK, CD79B and PI3K genes are positive regulators of this pathway. In order to validate these results and given the abundance of inhibitors of different proteins in the BCR pathway, we run a focused combination screen using MALT1 inhibitor MI-2 and inhibitors against other proteins in the pathway in 4 MALT1 sensitive cell lines. Combinations with PI3K inhibitors were most synergistic (combination index (CI) ranging 0.12-0.67), while BTK and PKC inhibitors showed an additive effect (CI ranging 0.7-0.9). These results were confirmed using a second MALT1 inhibitor, mepazine. In order to characterize the molecular mechanism by which MALT1 inhibition cooperates with PI3K, we focused on the FDA approved drug Idelalisib. In vitro treatment of cells with MI-2 and Idelalisib showed that effect on cell growth was a combination of decreased proliferation and increased apoptosis. Moreover, we found a decrease in AKT phosphorylation followed by a decrease in FOXO1 T24 phosphorylation and an accumulation of FOXO1 protein. This result correlates with our finding that FOXO1 knockdown favors MALT1 inhibition resistance. In vivo treatment of TMD8 xenografts with a combination of MI-2 and Idelalisib showed a stronger effect than either drug used as a single agent or vehicle, confirming the increased efficacy of the combination over either drug alone. In summary, we have used an shRNA library screening in order to determine which proteins and pathways cooperate with MALT1 inhibition to kill ABC-DLBCL and to evaluate combinatorial treatments in an unbiased manner. This same approach has pointed out TNFAIP3 and FOXO1 as possible biomarkers of response. This is especially interesting since these two proteins are mutated in a proportion of ABC-DLBCL patients and could affect response to treatment not only against MALT1 inhibitors but potentially any BCR targeted therapy. Disclosures Melnick: Janssen: Research Funding.

Aberrant activation of the B cell receptor (BCR) signaling network drives survival and proliferation of many B cell malignancies, such as activated B cell like diffuse large B cell lymphoma (ABC-DLBCL). A number of small molecule inhibitors targeting various kinases in the BCR signaling network have been developed. However, clinical application of these targeted agents is facing several challenges such as low response rate and acquired drug resistance. The limited efficacy of single agent targeted therapy is at least partially due to pathway reactivation through crosstalks and compensatory circuits. By simultaneously repressing multiple nodes in the signaling network, combination therapy has the potential to completely extinguish signaling and induce more potent and durable response. The complexity of BCR signaling network makes it difficult to infer which combinations will be effective and synergistic. Given the large number of possible drug combinations, comprehensive experimental screening – including exploration of multiple dosages – is not practically feasible. Computational models of signaling networks that can accurately reconstruct signaling dynamics in silico may represent a useful alternative to experimental screening and trial-and-error experimental investigation. Here, we developed a computational model that integrates signal transduction, tumor growth, and drug kinetics to accurately simulate BCR signaling dynamics and the effect of drug-induced perturbation on signaling output and cell viability. We used this model to predict effective combinatorial therapy in silicoand validated some of these predictions using in vitro experiments. Based on experimentally verified protein-protein interactions, we constructed the first detailed kinetic model of the BCR signaling network covering three major signaling pathways downstream of BCR, namely NFkB, PI3K/AKT and MAPK. The model captured complex crosstalk between these three pathways and multiple feedback loops. Simulated kinase activation time courses under temporal antigen stimulus successfully recapitulated normal BCR signaling dynamics as reported in literature. Using published drug response data in the BCR signaling-dependent ABC-DLBCL cell line TMD8, we trained a tumor growth model which in combination with the kinetic model enabled reliable prediction of viability response of many drug combinations at various dosages. For example, predicted viability response of BTK inhibitor ibrutinib in combination with inhibitors targeting other kinases in the network, e.g. BKM-120 against PI3K, sotrastaurin against PKC-beta closely matched previously published experimental data in TMD8 (r>0.86,p We then sought to identify synergistic drug combinations by simulating viability response at 10x10 virtual dosages for each drug combination and estimating synergism using Bliss independence model. Computational screening predicted dual blockage of LYN and SYK as the most synergistic combination, which we confirmed experimentally by treating TMD8 cells with LYN inhibitor Dasatinib and SYK inhibitor R406 at multiple doses. Finally we sought to use our model to predict biomarkers of sensitivity and resistance to specific treatment strategies. By integrating expression levels of BCR signaling network components assessed by published primary DLBCL RNA-seq data, we simulated patient-specific drug responses and computed the correlation between expression level of specific components with viability response under specific treatments. We found that overexpression of PTP1B, which dephosphorylates BTK substrate PLCg2, predicts relative sensitivity to BTK inhibition. Supporting this prediction, we observed increased PTP1B expression in DLBCL cell lines sensitive to ibrutinib treatment, suggesting PTP1B as potential biomarker for ibrutinib sensitivity. In summary, this study provides a novel approach to computationally optimize combinatorial targeted therapy against aberrant BCR signaling and paves the way for the discovery of effective patient-specific drug combinations. Disclosures Melnick: Bioreference: Scientific Advisory Board, Scientific Advisory Board Other; Calgene: Consultancy; Janssen: Research Funding; Genentech: Speakers Bureau.

Reprogramming of tumor cell metabolism contributes to disease aggressiveness and chemoresistance, but how this process is regulated on the molecular levels is unclear. Hsp90 regulates diverse cellular processes through its interaction with co-chaperones and client proteins. Although basally expressed, Hsp90 is almost universally overexpressed in malignant cells due to increased levels of internal (oxidants, genomic instability, unfolded proteins) and external (hypoxia, nutrients limitation, drugs) stressors. Indeed, Hsp90 is starting to be acknowledged also for its role as an integrator of nutrient sensing, tumor bioenergetics and metabolic stress response (Chae et al, Cancer Cell, 2012). Here we report that active cytosolic Hsp90 binds and stabilizes enzymes regulating central metabolic processes in DLBCL, and provides an actionable target for this disease. Despite its overexpression, not all DLBCL are equally responsive to Hsp90 inhibitors. Using a panel of 10 DLBCL cell lines we found that those with higher glycolytic activity were more likely to respond to the Hsp90 inhibitors 17-DMAG and PU-H71. Global metabolomics profiling (GC-MS and UPLC-MS) of Hsp90-dependent (OCI-Ly7, OCI-Ly1, SU-DHL6, OCI-Ly10) vs. Hsp90-independent (Karpas422) cells identified “purine metabolism” (p=0.000027) and “pyrimidine metabolism” (p=0.0017) as upregulated, whereas “TCA cycle” was downregulated (p=0.000027). We validated these findings by short-term (6 h) exposure of OCI-Ly1 and OCI-Ly7 cells with non-cytotoxic doses of PU-H71 and found a highly reduced oxygen consumption rate (OCR) as well as glycolytic rate (ECAR). This was accompanied by a sharp decrease in glucose consumption and lactate production and by only a slight reduction in glutamine consumption. Unexpectedly, the drop in glycolysis and respiration in OCI-Ly1 and OCI-Ly7 did not lead to a significant decrease either in ATP or reducing power, suggesting that actually both the production and the utilization of ATP and redox equivalents are shut down when Hsp90 activity (an ATP-hydrolyzing enzyme) is inhibited. To identify the molecular basis for these differences, we characterized the active Hsp90-associated proteome in DLBCL by using chemical affinity purification followed by MS in cytosolic fractions of two Hsp90-dependent DLBCL cell lines (OCI-Ly1 and OCI-Ly7) in duplicates. This method pulls down only active Hsp90 multichaperone complexes (Moulick et al, Nat Chem Biol, 2011). Pathway enrichment analysis identified “cellular metabolism” as a significant process actively chaperoned by Hsp90 (ratio of enrichment 4.35, p=8.22E-15). Specifically, “nucleotides” and “carbohydrates” metabolism were the two most represented pathways, suggesting their critical role in Hsp90-dependent DLBCL cells. To functionally validate these findings, we integrated the proteomics analysis with serum metabolomics obtained from 10 OCI-Ly7-xenografted mice treated with PU-H71 vs vehicle (75 mg/m2 for 24 h (n=5)). Bioinformatic analysis revealed that PU-H71-treated mice had significant changes in 122 metabolites including lower levels of xanthine, hypoxanthine, adenosine and inosine, suggesting an alteration of the purines metabolism. IMPDH1/2 is the enzyme catalyzing the first committed and rate-limiting step of de novo guanine nucleotide biosynthesis. To determine whether IMPDH stability depends on Hsp90 activity, we treated a panel of 6 DLBCL cell lines (including OCI-Ly7) with their GI50 for PU-H71 or 17-DMAG for up to 24 h and found a time-dependent decrease in IMPDH2 protein levels. Finally, SCID mice were xenografted with OCI-Ly1 and SU-DHL6 and treated with vehicle, PU-H71, mycophenolate mofetil (MMF, the pro-drug of the IMPDH inhibitor mycophenolic acid) and the combination of PU-H71 and MMF. We found that the combination of drugs exhibited a greater antitumor effect than each drug alone (p=0.002 for SU-DHL6 and P Altogether, our work supports a role for Hsp90 as a necessary component in a subset of DLBCLs to build-up the metabolic features that allow these tumor cells to meet the requirements for their unrestrained growth and proliferation. Disclosures No relevant conflicts of interest to declare.