Repurposing NAMPT Inhibitors for Germinal Center B Cell–Like Diffuse Large B-Cell Lymphoma Free

Diffuse large B-Cell lymphoma (DLBCL), the most common B-cell lymphoma, is curable by chemoimmunotherapy, but ∼30% of cases are resistant to treatment and remain an unmet clinical need (1). New approaches developed to tackle this need include cellular therapies such as chimeric antigen receptor T cells (2), bispecific antibodies (3), and small molecules targeting several signal transduction mediators, including PI3K (4) and exportin 1 (5). However, these strategies exhibit success in a discrete set of cases, likely reflecting the vast genetic heterogeneity of this disease (6). Thus, the search for new approaches is still warranted.

The classification of DLBCL in molecular subtypes has represented a rational strategy to narrow down actionable vulnerabilities. The gene expression–based cell-of-origin (COO) classification subdivides DLBCLs into germinal center B cell–like (GCB) and activated B cell–like (ABC) subtypes, which differ in transcriptional profile, genetic alterations, and responses to immunochemotherapy, with ABC-DLBCL having inferior prognosis (7). More recently, classifiers based on genetic features such as nonnegative matrix factorization (8) and LymphGen (9) have identified molecular subtypes characterized by coordinate recurring genetic features. Still, because heterogeneity persists within these subtypes, it remains difficult to precisely link genetic lesions to molecularly targeted therapies.

Emerging strategies include the repurposing of drugs approved or in advanced clinical development for other diseases (10, 11). The advantage of this strategy is three-fold. First, it allows to identify drugs that are specific for subsets of a disease and therefore unlikely to yield broad toxicities. Second, the annotation of the target of the active drugs can inform on pathway dependencies not readily identifiable based on genetic data. Finally, repositioning drugs already approved is a faster and less resource-intensive route to the clinic than de novo drug development. Notable examples of drug repositioning include thalidomide, which was originally developed as sedative and then repurposed for multiple myeloma in combination with dexamethasone (12), and imatinib, first approved for chronic myeloid leukemia (13) and then approved for gastrointestinal stromal tumors (14). Of note, imatinib was initially developed as an inhibitor of the Abelson kinase, whereas its activity in gastrointestinal stromal tumors is due to inhibition of c-KIT and platelet-derived growth factor receptor-α. This difference illustrates how repositioning may be classified as “on-target” if the target is the same for the original and new indication(s), or “off-target” when the new activity is explained by another mechanism (10). In the case of DLBCL, we have shown the “off-target” repositioning activity of dasatinib, a Src/Abelson kinase inhibitor approved for B-cell acute lymphoblastic leukemia and chronic myeloid leukemia, as a FYN inhibitor (10, 15).

Here, we provide preclinical evidence that a subset of GCB-DLBCL is sensitive to inhibition of the enzyme nicotinamide (NAM) phosphoribosyl transferase (NAMPT), which catalyzes the conversion of NAM to β-NAM mononucleotide (β-NMN). NAMPT catalyzes the rate-limiting step of the NAD salvage pathway, which is the main route for NAD regeneration and synthesis in mammals. Of note, NAD can be also synthetized from nicotinic acid via the Preiss–Handler pathway or de novo from tryptophan via the kynurenine pathway (16) and may also be salvaged from NAM riboside through the NAM riboside kinases 1 and 2. Besides its central role as a redox carrier in cellular energy metabolism, NAD also serves critical functions in regulating posttranslation modifications, including NAD-dependent deacetylation, which is mediated by the sirtuin family of proteins, and ADP-ribosylation, which is catalyzed by the PARP family (17). These processes involve the breakdown of NAD and the release of NAM that can re-enter the salvage pathway. NAD levels are therefore the result of the balance between the activities of the biosynthetic and consumption pathways. Here, we validated the preclinical efficacy of NAMPT inhibitors (NAMPTi) in vivo for a defined subset of GCB-DLBCL and identify potential biochemical and genetic biomarkers for the stratification of DLBCL cases sensitive to NAMPT inhibition.

For drug repurposing screening, we used a custom-assembled library of 211 compounds, including approved drugs as well as compounds in advanced clinical development for diseases other than DLBCL. The library is representative of multiple classes of molecules, including those targeting DNA damage response (p53, PARP), cell cycle (cyclin-dependent kinases and polo-like kinases), epigenetics activities (EZH2 and DOT1L), and the Notch and Sonic Hedgehog pathways (see Supplementary Table S1 for a fully annotated list). The library was tested in a high-throughput format on a panel of eight DLBCL cell lines including four ABC-DLBCL (OCI-LY10, RCK-8, HBL1, and OCI-LY3) and four GCB-DLBCL (SUDHL4, DB, WSU-NHL, and FARAGE). Each compound was tested on each cell line in a 48-hour dose–response assay with concentrations ranging from 20 µmol/L to 0.01 nmol/L in 1:3 serial dilutions and three to six replicates for each concentration point. The results were analyzed using a five-parameter logistic dose response curve for each assay, which was then used to compute the AUC. The DLBCL subtype specificity for each compound was determined by calculating the difference between the AUC means of the ABC versus GCB cell lines [Fig. 1A (y-axis)] and comparing it to the compound’s average activity, defined as the average AUC of the whole panel (x-axis).

This analysis identified the NAMPTi FK-866 as the lead compound for GCB specificity (Fig. 1A and B; Supplementary Table S1). Two additional, chemically distinct NAMPTis, STF-118804 and KPT-9274, showed similar specificity/activity ratio as FK-866, although at lower potency (Fig. 1A; the dotted green line indicates a 30% ratio between activity and specificity). NAMPT inhibition was comparable or better in terms of activity and subtype specificity than other classes of inhibitors currently in use or in clinical development for DLBCL and other B-cell malignancies, including PI3K and BTK inhibitors (IPI-145, acalabrutinib), inhibitors of nuclear export (selinexor), heat shock protein inhibitors (AUY922), and kinesin inhibitors (SB-743921; Fig. 1A; Supplementary Fig. S1; Supplementary Table S1).

To corroborate these findings, we tested an extended panel of 35 DLBCL cell lines (9 ABC and 25 GCB) for sensitivity to the three NAMPTis (FK-866, STF-118804, and KPT-9274) in 48-hour dose–response assays. GCB specificity was measured by the difference of IC₅₀ means between the two subtypes. The results showed that, although all GCB cell lines were generally more sensitive than ABC cell lines (see Fig. 1C for representative data with FK-866; P = 0.04), they displayed a broad range of sensitivities. In particular, GCB cell lines were classified based on FK866 sensitivity into GCB-resistant (GCB-R), defined as having z-scored IC₅₀ greater than 0.05, and GCB-sensitive (GCB-S), defined as having z-scored IC₅₀ equals or lower than 0.05 (Fig. 1C). Taken together, these results identify the NAD salvage pathway as a dependency of GCB-DLBCL and suggest that a subgroup of these tumors is particularly sensitive to its inhibition (see below).

We then tested the “on-target” versus “off-target” activity of NAMPTis, which is relevant because KPT-9274 was initially characterized as a PAK-4 inhibitor and subsequently redefined as NAMPTi (18). Toward this end, we first computed the correlation between the IC₅₀ values of each NAMPTi pair in the entire panel of DLBCL cell lines, which revealed a high degree of concordance as indicated by the R Pearson correlation coefficient (Fig. 2A). We then used three GCB-S cell lines to determine whether the activity of STF-118804, KPT-9274, and FK-866 could be rescued by addition of the NAMPT reaction product β-NMN. Consistent with on-target activity, β-NMN was able to rescue the activity of all three inhibitors over a broad range of doses in all three cell lines tested (Fig. 2B and C). The same cell lines were tested for response to each inhibitor after transduction with Venus-tagged vectors expressing NAMPT wild type (WT) or NAMPT H191R, a binding mutant that has been previously shown to confer cross resistance to several NAMPTis (19), along with empty vector (EV) as control. The results showed that reconstitution with the NAMPT H191R mutant was able to abrogate the cell sensitivity to all three inhibitors in all three cell lines, thus demonstrating that target engagement by these compounds is necessary for their activity (Fig. 2D and E). Finally, metabolomic profiling of WSU-DLCL2 cells treated for 16 hours with 1 µmol/L Karyopharm Therapeutics (KPT)-9724 showed NAD was the most depleted metabolite (89% reduction compared with control; Supplementary Fig. S2A–S2D; Supplementary Table S2), followed by NADPH and NAM. Notably, KPT treatment was also associated with decreases in the levels of metabolites derived from NADP-dependent reactions, such as the entry reactions in the pentose phosphate pathway 6-phosphogluconate and 2-hydroxyglutarate (generated by the NADPH-dependent isocitrate dehydrogenase 1 enzyme). Collectively, these data support the “on-target” activity of all three NAMPTis.

Sensitivity to NAD salvage inhibition has been ascribed to the balance of the competing synthetic and consuming activities (17). This notion predicts that an increase in synthesis or a decrease in consumption results in reduced sensitivity to NAD salvage inhibitors. Consistently, transduction of NAMPT WT cDNA in GCB lines was observed to decrease sensitivity to NAMPT inhibition in all three cell lines (Fig. 2D and E; compare NAMPT WT to EV AUCs). We thus hypothesized that differences in transcripts levels of synthesis and consumption enzymes may confer differences in sensitivities between the GCB-S and GCB-R subsets, and between GCB-DLBCL versus ABC-DLBCL.

Toward this end, we analyzed the transcript levels of the 13 biosynthetic and 22 consuming enzymes that are annotated for NAD synthesis and consumption (16, 20). We found that NAMPT and PARP1 were the most abundant transcripts across the whole panel [average transcripts per million (TPM): NAMPT = 47, PARP1 = 150; Supplementary Fig. S3; Supplementary Table S3]. Notably, NAMPT was the only NAD synthetic transcript that showed significantly higher levels in GCB-R versus GCB-S and in ABC versus GCB cell lines (Fig. 3A). In addition, analysis of DepMap data showed that NAMPT single-guide RNAs (sgRNA) were consistently depleted in DLBCL lines, whereas those against other NAD biosynthetic enzymes were not (Supplementary Table S4). Conversely, and consistent with our hypothesis, PARP1 levels were lower in ABC- versus GCB-DLBCL cell lines and were reduced in GCB-R versus GCB-S cells, although the differences did not reach statistical significance (Fig. 3B). Accordingly, the NAMPT:PARP1 (NP) mRNA ratio was significantly higher in ABC-DLBCL compared to GCB-DLBCL and in GCB-R compared to GCB-S DLBCL cell lines (Fig. 3C), thus correlating with the distinct sensitivities to NAMPT inhibition of the three DLBCL subsets. The differences in NP mRNA ratio between ABC-DLBCL and GCB-DLBCL were confirmed in two distinct publicly available cohorts of DLBCL primary cases (Supplementary Fig. S4A–S4F; refs. 7, 8).

In order to directly confirm whether the NP ratio was functionally involved in NAMPTi sensitivity versus resistance, we experimentally manipulated the level of these two enzymes and monitored for shifts in sensitivity to NAMPT inhibition. To this end, we transduced SUDHL4 cells (GCB-DLBCL) with Venus-tagged NAMPT WT and H191R cDNA alongside an empty vector control, sorted high and low expressing fractions, and performed dose–response assays to the NAMPTi KPT-9274 (Fig. 3D–F). The NAMPT-WT-low transduced fraction was significantly less resistant than the NAMPT-WT-high fraction, whereas, as expected, no difference was observed between the two fractions of the NAMPT H191R mutant and the two EV fractions, indicating that increasing NAMPT levels induces resistance in a dosage-dependent manner (Fig. 3E and F). To then test whether PARP1 activity modulates NAMPTi sensitivity, we derived isogenic pools of two GCB-S cell lines (KARPAS-422 and WSU-DLCL2) carrying two independent sgRNAs targeting PARP1 (PARP1 #1 and PARP1 #2) or a neutral control region (21) and measured their response to KPT-9274. Both sgRNA sequences were able to abolish PARP1 protein in both cell lines (Fig. 3G) and induced 10 to 100 fold resistance to KPT-9274 compared with the control sgRNAs (Fig. 3H and I). Blockade of PARP activity by the PARP inhibitor rucaparib also determined a dose-dependent shift to resistance in both lines, as shown by KPT-9274 response in the presence of increasing concentrations of rucaparib (Fig. 3J–M). Altogether, these data support NAMPT and PARP1 activity as determinants of sensitivity to NAMPTi and identify the NP mRNA ratio as a measure to predict response to NAMPT inhibition.

Next, we asked whether specific genetic lesions in DLBCL are associated with favorable NP ratios in GCB-DLBCL cases, as this would provide potential biomarkers to identify those patients most likely to benefit from NAMPTi therapy. To this end, we analyzed a publicly available dataset of 482 DLBCL cases (138 GCB-DLBCL) characterized by RNA-seq, exome sequencing, and annotation for BCL2 and BCL6 structural variants (22). We assembled a list of 119 genetic features that comprise all the lesions significantly associated with any of the LymphGen subtypes (with the exception of PDCD1LG2/CD274 translocation and copy number–based features that were not available). We then calculated the ratio between the average NP ratio of samples harboring the genetic lesion and the average NP ratio of samples in which the same lesion is absent, for 36 lesions found in more than 10% of cases. Individually, none of these alterations were significantly associated with a lower NP ratio; however, cases carrying EZB-associated alterations showed the lowest values (Supplementary Table S5). In particular, we noted that cases carrying the EZB hallmark BCL2 translocation without the concurring presence of BCL6 and/or MYC rearrangements (n = 39) showed a significantly lower NP ratio than unrearranged GCB cases (n = 94; P < 0.01; Student t test with Welch correction; Fig. 4A; Supplementary Table S5). Consistently, 56/121 DLBCL cases classified in the lower NP quartile were GCB-DLBCL (P < 0.0001; Fisher exact test), of which 20 carried BCL2 translocations (Fig. 4B). Analogously, BCL2-translocated cases were significantly enriched in the lower NP quartile (n = 20/121; P = 0.001; Supplementary Table S6). Taken together, these findings indicate that BCL2-translocated GCB-DLBCL might be particularly vulnerable to NAMPTis. Interestingly, inhibition of NAMPT by KPT-9274 significantly enhanced the detrimental effect of CRISPR/Cas9-mediated BCL2 genetic deletion in two BCL2-translocated GCB-DLBCL lines (Fig. 4C and D) and was synergistic with pharmacologic blockade of BCL2 by venetoclax (Fig. 4E). These results identify BCL2 translocations as a potential clinical biomarker for NAMPTi sensitivity in GCB-DLBCL.

To examine the activity of NAMPTis in vivo, we established luciferized xenotransplants of the BCL2-translocated GCB cell line KARPAS-422 in NOD/SCID mice and treated them with KPT-9724 (oral administration, 150 mg/kg/day for 28 days). This treatment showed a robust suppression of xenotransplant growth in vivo as measured by luciferase imaging (photons/seconds/cm²; Supplementary Fig. S5A and S5B) and tumor volume (Supplementary Fig. S5C).

To preliminarily validate BCL2 status as a biomarker for NAMPTi therapy, we searched the patient-derived xenograft (PDX) PRoXe database (23) for DLBCL PDXs classified as GCB and with available transcriptional profiles as well as BCL2 translocation status. We retrieved four BCL2-translocated and one BCL2 WT GCB-DLBCL PDX. Consistent with our previous observations, all BCL2-translocated PDXs showed low NP ratios that were comprised between 0.10 and 0.18, compared with 1.35 in the BCL2 WT PDX (Fig. 4F). Treatment of the BCL2-positive PDX #20954 with KPT-9274 (n = 4 mice, vs. 4 control-treated) led to a three-fold reduction in tumor volume at day 21. In contrast, responses in the BCL2 WT #75549 PDX were markedly inferior and not significantly different from vehicle treatment (P = 0.11; Fig. 4G). These data are consistent with the potential utility of BCL2 translocation status in predicting response to KPT-9274.

Although several treatment approaches have emerged for the prognostically negative ABC-DLBCL subtype, therapeutic options for the GCB subtype are lacking despite the presence of a significant subset of high-risk GCB-DLBCL (24). In this work, we propose NAMPT as a robust therapeutic target for a subset of GCB-DLBCL. Our results also identify the NP mRNA ratio as a general determinant of sensitivity to NAMPT inhibition and provide preclinical evidence that GCB-DLBCL carrying BCL2 translocations are the most sensitive DLBCL subgroup.

Several NAMPTis have reached the clinical development stage, were proven safe, and showed some level of activity in a variety of cancer types (16). In the context of B-cell malignancies, FK-866 was shown to have preclinical activity in several diseases including Waldenström macroglobulinemia in combination with ibrutinib (25), chronic lymphocytic leukemia (CLL) in combination with the anti-CD20 antibody rituximab, and Burkitt lymphoma cell lines (26). The NAMPTi KPT-9274, which is extensively used in our study, is an orally bioavailable drug that has been demonstrated safe and has shown preliminary activity in follicular lymphoma and in a few uncharacterized DLBCL models (27). To date, KPT-9724 seems as the most promising compound for future clinical development. Our biochemical and genetic rescue approaches show that KPT-9274, as well as the other NAMPTis, act “on-target” in DLBCL, suggesting the possibility that other components of the NAD salvage pathway may represent testable therapeutic targets.

Our findings provide insights on how the sensitivity spectrum across DLBCLs depends on the balance between synthetic and degradative activities. The NP transcript ratio represents a faithful indication of the levels of sensitivity to NAMPT inhibition, a high ratio being associated with ABC-DLBCL in both cell lines and primary cases, and a low ratio being associated with GCB-DLBCL cases. In addition, the NP transcript ratio was able to distinguish the more sensitive subset of GCB-DLBCL. Although the NP ratio mechanistically represents the most valid biomarker of sensitivity, its utilization in a clinical context has practical limitations and may require further calibration in order to define clinically valid thresholds for the identification of sensitive cases.

Our data point to high NP transcript ratio as an indicator of resistance. Compensation of NAMPTi activity by other NAD synthetic pathways has been shown in other preclinical contexts (28) and NAPRT activity has been connected with NAMPTi resistance (28). Although we did not detect activation of the NAD biosynthetic pathway in resistant models by metabolic profiling, we cannot exclude that this could occur as a mechanism of adaptive resistance in patients treated with NAMPTi. Analysis of clinical cohorts will be helpful to determine whether resistance develops and if it is driven by compensation in NAD synthesis.

Finally, our search for genetic hallmarks correlating with minimal NP ratio within the GCB subtype led to the identification of cases carrying the BCL2 translocation (in the absence of MYC or BCL6 genetic alterations) as the ones with the lowest NP ratio. Although the BCL2 translocation status, as a biomarker for NAMPTi response, did not detect all sensitive cases identified by NP ratio alone, it has the advantages of offering high fidelity and being an established diagnostic test, thus not requiring the development of a companion diagnostic assay. Consistently, functional cosuppression of BCL2 and NAMPT showed potent effect in BCL2-translocated GCB-DLBCL both in vitro and in vivo, although tested in a limited number of models. Further studies will be needed to investigate whether and how low NP ratio and BCL2 translocation are causally linked.

In conclusion, the work presented here supports that NAMPT inhibition will be beneficial especially to the GCB-DLBCL population and potentially to transcriptionally similar transformed follicular lymphoma cases (29), and that assessment of BCL2 translocation status may represent an effective biomarkers for patient selection for NAMPTi treatment, with the potential to address an unmet clinical need in the GCB-DLBCL subtype.

The OCI-LY-4 (RRID: CVCL_8801) and OCI-LY-10 (RRID: CVCL_8795) DLBCL cell lines were grown in Iscove's modified Dulbecco's medium (IMDM) supplemented with 55 µmol/L β-mercaptoethanol (Sigma-Aldrich), 20% human plasma (New York Blood Bank), and 1% penicillin/streptomycin (Thermo Fisher Scientific, #MT30-002). All other 32 DLBCL cell lines used in this study were grown in IMDM supplemented with 10% FBS and 1% penicillin/streptomycin. These include BJAB (RRID: CVCL_5711), DB (RRID: CVCL_1168), DOHH2 (RRID: CVCL_1179), FARAGE (RRID: CVCL_3302), HBL1 (RRID: CVCL_4213), HT (RRID: CVCL_1290), KARPAS-422 (RRID: CVCL_1325), NU-DHL-1 (RRID: CVCL_1876), OCI-LY-1 (RRID: CVCL_1879), OCI-LY-18 (RRID: CVCL_1880), OCI-LY-19 (RRID: CVCL_1878), OCI-LY-3 (RRID: CVCL_8800), OCI-LY-7 (RRID: CVCL_1881), OCI-LY-8 (RRID: CVCL_8803), PFEIFFER (RRID: CVCL_3326), RCK8 (RRID: CVCL_1883), RIVA (RRID: CVCL_1885), RL (RRID: CVCL_1660), SU-DHL-10 (RRID: CVCL_1889), SU-DHL-16 (RRID: CVCL_1890), SU-DHL-2 (RRID: CVCL_9550), SU-DHL-4 (RRID: CVCL_0539), SU-DHL-5 (RRID: CVCL_1735), SU-DHL-6 (RRID: CVCL_2206), TMD8 (RRID: CVCL_A442), TOLEDO (RRID: CVCL_3611), U2932 (RRID: CVCL_1896), WSU-DLCL2 (RRID: CVCL_1902), WSU-NHL (RRID: CVCL_1793), HLY1 (RRID: CVCL_H207), LIB (RRID: CVCL_H209), and MIEU (RRID: CVCL_H208), the latter three being a gift of Dr. Brousset and Dr. Delsol (University of Toulouse). β-NMN was purchased from Sigma (#N3501). Cells were tested for Mycoplasma contamination using the MycoSensor PCR Assay Kit (Agilent Technologies) and were used in log phase of growth. The identity of all DLBCL cell lines was verified by short tandem repeat profiling and/or by analysis of known somatic single-nucleotide variants (30).

Eight DLBCL cell lines (four ABC: OCI-LY-10, OCI-LY-3, HBL1, and RCK-8; and four GCB: SUDHL4, WSU-NHL, DB, and FARAGE) were screened in 384-well format using a custom-assembled library of 211 compounds, including approved drugs and compounds in advanced clinical development for diseases other than DLBCL. Cells were seeded at 1,000/well in 50 µL media and compounds applied after 24 hours. Viability was determined by CellTiter-Glo (Promega) readings after 48 hours drug exposure. IC₅₀ values and AUCs were assessed using five-point or four-point logistic regression curves by the software package nplr (https://cran.r-project.org/web/packages/nplr/index.html). The full list of the library drugs is reported in Supplementary Table S1.

Thousand cells per well were plated in 384-well plates in 50 μL media using a Matrix WellMate dispenser and treated with the indicated concentrations of the NAMPTis FK-866 (Selleck) and STF-118804 (Sigma), with or without venetoclax or rucaparib (Selleck). KPT-9274 was provided by Karyopharm. Compounds were resuspended in DMSO and pinned using a HP D300e digital dispenser. Treated wells were normalized to the highest level of DMSO used across the plate. Dose–response assays were performed with at least four replicates for concentration point. CellTiter-Glo (Promega) readings were used to assess viability. Curve fitting was performed using nplr as described above or GraphPad five-point logistic functions. Error bars represent SDs. For combination assays, cells were seeded in 384-well plates at 1,000 cell/well as described above. Drugs were pinned in a 7 × 7 matrix in four replicates alongside single-agent treatments on the same 384-well plate. Concentration ranges were between 20 and 1 μmol/L for venetoclax and between 10 and 0.1 μmol/L for KPT-9274. EC₅₀ values were then calculated for single-agent treatments and for the cotreatment conditions, and combinatorial indexes were calculated according to the method of Chou and Talay (31).

As a tool to further dissect the factors influencing sensitivity to NAMPTi, we defined cells as sensitive or resistant by calculating the z-score of the IC₅₀ values of each compound across all GCB-DLBCL cell lines; GCB lines with z-score greater than 0.05 were arbitrarily classified as GCB-R, whereas those with z-score lower than 0.05 were classified as GCB-S.

To generate NAMPT-P2A-VENUS constructs, a ubiquitin C (UBC)-driven constitutive lentiviral backbone was first derived by removing inducible components and rTTA3 from the pINDUCER 11 (a gift from SJ Elledge). V5-tagged NAMPT cDNA was cloned in NotI/BglII sites by removing the NAMPT stop codon in frame to a P2A-VENUS sequence. The NAMPT H191R mutant was obtained by site-directed mutagenesis (Quick Change II, Agilent Technologies). The cDNA encoding for the Luciferase–Tomato fusion protein (RRID: Addgene_32904) was cloned in the same UBC-driven backbone by means of NotI/MluI restriction sites.

Lentiviruses were produced by calcium phosphate cotransfection of 293T cells with the packaging vector PSPAX2 (a gift from Didier Trono; RRID: Addgene_12260) and pMD2.G (RRID: Addgene_12259) as previously described (16). Sixteen hours after transfection, the media were replaced with 6 mL IMDM. Viral supernatants were collected at 40 hours after transfection, filtered using a 0.45 µmol/L syringe filter, and supplemented with 4 µg/mL polybrene (Sigma-Aldrich). One hundred thousand target cells were spinoculated with 2 mL of virus supernatant for 90 minutes at 1,900 rpm. Infection was repeated after 12 hours using a second viral collection. Infected cells were resuspended in fresh media, grown for additional 24 hours, and sorted according to the tagged fluorescent protein encoded in the vector by using a BD Influx cell sorter prior to expansion for subsequent experiments.

WSU-DLCL2 and KARPAS-422 were lentivirally transduced as described above with a vector expressing Streptococcus pyogenes CAS9 and blasticidin resistance (a gift from Feng Zhang; Addgene plasmid #52962; http://n2t.net/addgene:52962; RRID: Addgene_52962). Cells were selected and expanded in media supplemented with 10 μg/mL blasticidin (Thermo Fisher Scientific, #R21001) for 10 days. The following sgRNAs were cloned in the pLKO5.sgRNA.EFS.GFP (kind gift from Dr. Benjamin Ebert; Addgene plasmid #57822; http://n2t.net/addgene:57822; RRID: Addgene_57822; ref. 32): P#1 (PARP#1), 5-GATGTCCACCAGGCCAAGGG-3; P#2 (PARP#2), 5-GATGGAAAAGTCCCACAC-3; PCNA, 5-GTCGAAGCCCTCAGACCGCA-3; BCL2#1, 5-CGCGGGGACGCTTTGCCA-3; BCL2#2, 5-GGCGATGTTGTCCACCAG-3; BCL2#3, 5-ACCCCACCGAACTCAAAGA-3; BCL2#4, 5-TCGCAGAGGGGCTACGAGT-3. All target sequences were derived from the VBC database (33). The control sgRNA (5-CCAGCGAGTGAAGACGGCAT-3) targets a neutral region in the PPP1R12C intron 1 that has been previously described (21). Infections were performed as described above. GFP-positive pools were sorted using a BD Influx sorter and expanded for subsequent experiments.

GFP-sgRNA populations (neutral vs. target gene-specific) were mixed 1:1 with untransduced CAS9-expressing cells and seeded in round-bottom 96-well plates at 10,000 cells/well in eight replicate wells per each sgRNA tested. The ratio between GFP⁺ and GFP⁻ cells was assessed at day 0 by means of an Attune Flow Cytometer (Thermo Fisher Scientific). Four wells were left untreated, whereas the remaining four were treated with 500 nmol/L KPT-9274. The GFP fraction was then assessed at day 3 after treatment in treated and untreated wells. The normalized GFP fraction was calculated by dividing for the corresponding day 0 value and then for the GFP fraction of the sg-neutral control under the same treatment condition.

Five million cells per sample were washed in cold PBS and lysed in 1% Triton-X lysis Buffer [150 mmol/L NaCl, 1 mmol/L EDTA, 1 mmol/L EGTA, 20 mmol/L Tris pH 6.8, supplemented with Protease Inhibitor Cocktail (Sigma), 1 mmol/L NaF, 1 mmol/L β-glycerophosphate, and 1 mmol/L phenylmethylsulfonylfluoride]. Protein concentration was quantified using the Bradford protein assay (Bio-Rad, #5000006), and 20 μg per sample were subjected to SDS-PAGE and transferred to nitrocellulose membranes (GE, #10600003) according to standard protocols. The following primary antibodies were used: NAMPT (Santa Cruz Biotechnology, #sc-393444, RRID: AB_2894708), Bcl2 (BD Pharmingen, #51-6511GR, RRID: AB_394048), α-tubulin (Sigma, #T5168, RRID: AB_477579), V5-tag (Thermo Fisher Scientific, #R96025, RRID: AB_2556564), PARP1 (Santa Cruz Biotechnology, #sc8007, RRID: AB_628105), and vinculin (Sigma-Aldrich SAB4200080, RRID: AB_10604160).

WSU-DLCL2 cells were treated with 1 µmol/L KPT-9724 for 16 hours, washed three times with ice-cold PBS, and frozen at −80°C. Samples were processed at the University of Colorado School of Medicine Metabolomics Core, and metabolomic analyses were performed by mass spectrometry, as described in detail in (34). Thawed samples were resuspended in chilled extraction solution [methanol:acetonitrile:water (5:3:2 v/v)] at 2 × 10⁶ cells per mL of extraction solution. After vortexing for 30 minutes at 4°C, samples were centrifuged at 12,000 g for 10 minutes at 4°C and supernatants isolated for metabolomics analyses. A volume of 10 μL of sample extract was run through a Kinetex C18 1.7 μm, 100 × 2.1 mm (Phenomenex) reversed phase column (positive ion mode—phase A: water, 0.1% formic acid; B: acetonitrile, 0.1% formic acid; negative ion mode—phase A: 1 mmol/L NH4Ac 95:5 water:acetonitrile; phase B: 1 mmol/L NH4Ac 95:5 acetonitrile:water) on a 5-minute gradient on an ultra-high pressure liquid chromatography system (Vanquish, Thermo Fisher Scientific). The ultra-high pressure liquid chromatography system was coupled in line with a high-resolution quadrupole Orbitrap instrument run in both polarity modes (QExactive, Thermo Fisher Scientific) at a resolution of 70,000 (at 200 m/z). The mass spectrometer operated either in positive or negative ion mode in separate runs. To quantify metabolite peaks, the raw files (.raw) containing information of the spectra of metabolites were converted to .mzXML using RawConverter. Metabolites were then assigned names using the Kyoto Encyclopedia of Genes and Genomes database by Maven software and in house standard libraries.

To measure the antitumor activity of KPT-9274 in vivo, KARPAS-422 cells were transduced with a lentiviral Luciferase–Tomato encoding vector and the Tomato-positive fraction was sorted and expanded as described above. Cells were diluted in PBS, mixed 1:1 with Matrigel (Corning, #356230), and injected subcutaneously in NOD/SCID mice at 0.5 × 10⁶ cells per mouse. Recipient mice were monitored by palpation and by bioluminescence imaging every 7 days. For imaging, mice were injected with XenoLight D-Luciferin (15 mg/mL in PBS, Perkin Elmer) at a 150 mg/kg dosage, anesthetized with isoflurane, and imaged with an IVIS Spectrum Optical Imaging System. Only mice carrying tumors that were detectable by caliper measurements (>2 mm diameter) and with total flux greater than 1 × 10⁷ luc photons were used for treatment studies. Volumes were calculated with the formula l/2 × w². KPT-9274 was administered by oral gavage as a 29.22% formulation alongside matching placebo. PDX mice were retrieved from the PRoXe collection, expanded in vivo by subcutaneous injection in NOD/SCID gamma mice (Jackson Laboratory) at 1 × 10⁷ cells per mouse, and subcutaneously transplanted for treatment studies at 5 × 10⁶ cells/mouse. All animal experiments were approved by the Institutional Animal Care and Use Committee of Columbia University. Animals were randomly assigned to vehicle control or drug treatment group.

RNA sequencing (RNA-seq) profiles for the 29 DLBCL cell lines have been described (GSE207388; ref. 35). For PDX profiles, paired-end RNA-seq data of the five PDX models was downloaded from the Sequence Read Archive (SRA) using SRA-Toolkit and their respective SRA ids (SRR5419313, SRR5419314, SRR5419315, SRR5419317, and SRR5419323). STAR aligner (v2.7.2a; ref. 36) was used to align sequencing reads to the human reference genome hg19. FeatureCounts (v.1.6.3; ref. 37) was used to generate raw gene expression counts, which were further normalized to TPM using Python script. We restricted the transcriptome to exon only and removed read-through genes, miRNA, rRNA, and antisense elements from the GENCODE GTF file (version 34). TPM expression of PDX models was merged with TPM expression of 29 DLBCL cell lines, and batch effect was removed using ComBat function in R (38). For the DLBCL cases, the NCI dataset (9) and linked clinical information were retrieved and processed as described (39). The DFCI dataset (8) was retrieved from the Gene Expression Omnibus (GEO) database (Accession No. GSE98588).

A list of 126 genes associated with the six LymphGen classes was compiled based on Wright and colleagues (22). After removing genes exclusively targeted by copy number aberrations (NFKBIZ, ING1, CNPY3, MIR17HG, REL, and SPIB) and PCDC1LG2/CD274, for which fusion status was not available, the remaining features were retained if observed at more than 10% frequency in the cohort. For each feature, samples were classified as mutated or WT on the basis of the mutational status, and average NP ratios were calculated for the two resulting groups.

Raw RNA-seq data from DLBCL cell lines can be obtained from the GEO database (Accession No. GSE207388). Raw RNA-seq data from the PDX models were downloaded from the SRA (SRR5419313, SRR5419314, SRR5419315, SRR5419317, and SRR5419323). Genomic data from the DLBCL NCI dataset were downloaded from the Database of Genotypes and Phenotypes (phs001444.v2.p1). Genomic data from the DLBCL DFCI dataset were retrieved from the GEO database (Accession No. GSE98588).

C. Scuoppo reports being an employee of Sapience Therapeutics. His current work is unrelated to the work reported here. No disclosures were reported by the other authors.

One of the Editors-in-Chief of Blood Cancer Discovery is an author on this article. In keeping with the AACR editorial policy, the peer review of this submission was managed by a member of Blood Cancer Discovery’s Board of Scientific Editors, who rendered the final decision concerning acceptability.

C. Scuoppo: Conceptualization, formal analysis, investigation, visualization, writing–original draft, writing–review and editing. B. Cai: Investigation. K. Ofori: Formal analysis, investigation. H. Scholze: Formal analysis, investigation. R. Kumar: Software, formal analysis. A. D’Alessandro: Resources, formal analysis, methodology. K. Basso: Data curation, writing–review and editing. L. Pasqualucci: Resources, supervision, investigation, project administration, writing–review and editing. R. Dalla-Favera: Conceptualization, resources, supervision, funding acquisition, writing–original draft, project administration, writing–review and editing.

We thank Dr. Charles Karan, Scientific Director of the High-Throughput Screening Facility at the JP Sulzberger Columbia Genome Center, for assistance with the repurposing drug screening, and the University of Colorado School of Medicine Metabolomics Core for their contribution to this manuscript. This work was in part supported by a Leukemia Lymphoma Society Translational Research Program Grant, the NIH (R35CA-210105 to R. Dalla-Favera), and NIH/NCI Cancer Center Support Grant P30CA013696. This research used the resources of the Herbert Irving Comprehensive Cancer Center; the Flow Core Facility supported in part by the Office of the Director, NIH under award S10OD020056. C. Scuoppo was supported by a MSKCC Lymphoma SPORE DRP award (P50CA-192937) and by an American Society of Hematology Restart Award. L. Pasqualucci was on leave from the University of Perugia. The DLBCL NCI dataset was accessed through the NIH database for Genotypes and Phenotypes (Genomic Variation in Diffuse Large B-Cell Lymphomas study, supported by the Intramural Research Program of the NCI, NIH, Department of Health and Human Services).