Schlafen 11 (SLFN11) Kills Cancer Cells Undergoing Unscheduled Re-replication

Schlafen 11 (SLFN11), a putative DNA/RNA helicase and endonuclease (1), is a growing focus of precision medicine because its expression is highly correlated with sensitivity to a broad range of DNA-targeting anticancer drugs across cancer cells (2, 3). Known SLFN11-dependent drugs include platinum derivatives (cisplatin and carboplatin), topoisomerase I (TOP1) inhibitors (camptothecin, irinotecan, topotecan, exatecan, and the indenoisoquinoline non-camptothecins, LMP400, LMP776, LMP744 and LMP517), topoisomerase II (TOP2) inhibitors (doxorubicin, daunorubicin, mitoxantrone, etoposide, and teniposide) and DNA synthesis inhibitors (gemcitabine, cytarabine, hydroxyurea and nucleoside analogs; refs. 3–5) as well as PARP inhibitors (olaparib and talazoparib; ref. 6). The causality of SLFN11 as a drug response determinant has been validated in isogenic SLFN11-proficient and -deficient cells regardless of the origins of tissues and by independent research groups (3, 6–10). The findings in cancer cell lines have been translated clinically in studies reporting better clinical response to DNA-targeting drugs in high SLFN11-expressing cancers than in low SLFN11-expressing cancers from different organs, including the breast (11, 12), ovary (12), stomach (13), bladder (14), lung (12), esophagus (15), medulloblastoma (16), and prostate (17).

Although the primary mechanisms of action for these SLFN11-dependent DNA-targeting agents are different, all ultimately induce replication stress (1, 18, 19). Typically, replication stress is marked by uncoupling between the replicative helicase complex (MCM2–7) and the replicative DNA polymerases, generating single-strand DNA gaps where replication protein A (RPA) filaments accumulate. Hence, forming extended RPA filaments on chromatin is one of the molecular characteristics of replication stress. Extended RPA filaments recruit and activate ataxia telangiectasia and Rad3-related protein kinase (ATR) with phosphorylation of RPA2 (S4/S8) and CHEK1 (S345), which are hallmarks of replication stress (4).

SLFN11 is also recruited to the RPA filaments at stressed replication forks where it induces lethal replication block (7, 20), which explains at least in part why SLFN11-proficient cells are more sensitive to DNA-targeting agents than SLFN11-negative cells. Other mechanisms of action of SLFN11 driving drug hypersensitivity include degradation of the replication initiation factor CDT1 (19), tRNA-cleavage leading to insufficient ATR synthesis (21), chromatin remodeling (22), activation of immediate early genes (JUN, FOS, p21, etc.; ref. 22), degradation of reversed replication forks (8), and protection from proteotoxic stress (23). Notably, our group has shown that the ATPase domain of SLFN11 is not required for the recruitment of SLFN11 to chromatin but for SLFN11 to block replication and kill cancer cells (20). More information on SLFN11 is available in recent reviews (1, 4). The detailed structure of SLFN11 has recently been revealed by cryo-electron microscopy analysis (24).

Re-replication is an aberrant process characterized by uncontrolled, continuous reinitiation of DNA synthesis within a given S-phase (25). Re-replication can be pharmacologically driven by inhibition neddylation with MLN-4924 (pevonedistat), which leads to the supraphysiologic and unscheduled accumulation of CDT1, an essential factor for replication initiation and timing. Hence, pevonedistat acts as an anticancer agent at least partly through enforced re-replication (26), and 41 clinical studies have been conducted and are ongoing (clinicaltrials.gov). The most advanced phase III study (No. NCT03268954) indicates improved overall survival for the pevonedistat+azacitidine group compared to the azacitidine-alone group in high-risk myelodysplastic syndrome (27). Currently, biomarkers for pevonedistat are not available in clinical settings.

In this study, we ran cell viability screens using 1,978 high-value small molecules known to modulate oncology targets, and pathways together with the National Center for Advancing Translational Sciences (NCATS; ref. 28) in two isogenic SLFN11-proficient and -deficient cell lines. This led to the discovery that MLN-4924 is a novel SLFN11-dependent drug and to the examination of the molecular mechanisms of how SLFN11 sensitizes cancer cells to MLN-4924.

DU145, CCRF-CEM, and K562 cell lines were obtained from the Developmental Therapeutics Program (RRID:SCR_003057), NCI/NIH and grown in RPMI1640 medium (1x, Gibco, 11875–093) added with 10% FBS (Gemini, 100–106) and 1% penicillin–streptomycin (Gibco, 15140–122) at 37°C in 5% CO₂. Gene modification was performed previously; the information can be found in our previous publications (6, 20). Wild-type (WT) and mutant DT40 cells were obtained from Dr. Shunichi Takeda (Kyoto University, Graduate School of Medicine, Sakyo-Ku, Kyoto, Japan). The details of the cells are summarized in our previous article (29). Drug libraries (MIPE Library, v4, Supplementary Tables S1 and S2) and small compounds (camptothecin, MLN-4924, and AHPN/CD437) were provided by the Developmental Therapeutics Program (NCI/NIH). TAS4464 was supplied by Taiho Pharmaceutical Co., Ltd. (Tsukuba, Ibaraki, Japan). All cell lines tested negative for mycoplasma contamination. Chemical structures of the 1,978 compounds (MIPE Library, v4) are provided as the Simplified Molecular Input Line Entry System (SMILES) form in our previous publication (30). The chemical structures and the synthesis methods for TAS4464 are available in Supplementary Fig. S1 (31). The chemical structures of CD437 (32) and MLN-4924 (33) are available in previous publications.

The anti-SLFN11 antibody (sc-515071X, 2 mg/mL, mouse monoclonal, Santacruz), the anti-total Chk1 antibody (sc-8408, mouse, Santacruz), the anti-phospho (S345)-CHK1 (2348S, rabbit monoclonal, Cell Signaling), the anti-phospho (S4/S8)-RPA2 (A300–245A), rabbit polyclonal, BETHYL) for immunofluorescence, the anti-RPA2 (Ab-3) (NL19L-100UG, mouse monoclonal, Oncogene) for immunofluorescence, the anti-RPA2 antibody (35869S, rabbit monoclonal, Cell Signaling) for Western blotting, the anti-phospho (S4/S8)-RPA2 (54762S, rabbit monoclonal, Cell Signaling) for Western blotting, the anti-phospho (S139)-H2AX (JBW301) (05–636, mouse monoclonal, Millipore), the anti–pan-Actin antibody (12748S, rabbit monoclonal, Cell Signaling), the anti-BrdU (B44) (347580, mouse monoclonal, Becton Dickinson), and the anti-CDT1 (D10F11) (8064, rabbit monoclonal, Cell Signaling) were used.

Each of the 1,948 compounds of the MIPE v4 library was tested in an 11-point dose response starting at 46 μmol/L with serial one to three dilutions. Briefly, cells were seeded into polystyrene, tissue-culture-treated 1,536-well white solid-bottom plates as previously reported (28). Seeding was performed with a Multidrop Combi Reagent dispenser (ThermoFisher) with a miniature 8-pin cassette at a density of 500 cells per well in 5-μL final volume. Immediately after dispensing the cells, 23 nL of compound stocks was transferred to individual wells using a Kalypsys Pin Tool. The plates were then covered with stainless steel Kalypsys lids and incubated at 37°C with 5% CO₂ and 95% relative humidity for 2 days. Forty-eight hours after the addition of the compound, 3 μL of CellTiter-Glo reagent assay (Promega) was added to each well using a Multidrop Combi Reagent dispenser (ThermoFisher). Plates were incubated for 15 minutes at room temperature with the stainless-steel lid in place, and relative luminescence units (RLU) were quantified using a ViewLux imager (PerkinElmer) with a 2′ exposure time. RLU for each well were normalized to the median RLU from the DMSO control wells as 100% viability. The half-maximal activity concentration (LAC₅₀) values were calculated automatically from the fitting curves.

Curve response class (CRC) classification originates from a dose–response high-throughput screen, in which normalized data were fitted to four-parameter dose–response curves using a custom grid-based algorithm to generate a CRC score for each compound dose response. CRC values of −1.1, −1.2, −2.1, and −2.2 are considered high-quality hits; CRC values of −1.3, −1.4, −2.3, −2.4, and −3 are inconclusive hits; and a CRC value of 4 is for inactive compounds (28).

For the viability assay in Fig. 1C, 2,000 cells were seeded, and the drug-containing medium was added in 384-well white plates in the final 40 μL of medium per well. Cellular viability was determined using the ATPlite 1-step kits (PerkinElmer). The ATP level in untreated cells was defined as 100%. The viability (%) of treated cells was defined as the ATP level of treated cells/ATP level of untreated cells × 100.

A total of 200 DT40 cells were seeded into each well of 384-well white plates, with 20 μL of culture medium used for the seeding process. Following the cell seeding, an additional 20 μL of culture medium containing drugs was added to each well. Plates were incubated at 37°C for 72 hours. Cell survival was determined using the ATPlite 1-step kit (Perkin Elmer Life Sciences). All procedures were performed in triplicate. To evaluate the relative cellular sensitivity of each mutant to WT cells, sensitivity curves were drawn by setting the survival of untreated cells as 100%. IC₉₀ values for each drug and cell line were determined as the crossing points between the 10% viability line and the survival curve connecting average points for each drug concentration. The IC₉₀ of each mutant was divided by the IC₉₀ of WT cells that were cultured on the same plate, and then the quotient was converted into a logarithmic scale (base 2). Each score was plotted on the same bar graph (29).

Whole-cell was lysed with RIPA lysis buffer. Samples were mixed with tris-glycine SDS sample buffer (Nacalai Tesque) and loaded onto tris-glycine gels (BioRad). Blotted membranes were blocked with 4% bovine serum albumin (BSA) (Sigma-Aldrich, A9418) in phosphate-buffered saline (PBS) with 0.1% tween-20 (PBST). The primary antibody was diluted in 4% BSA/PBST by 1:3000 for proteins other than anti-Histone H3 and anti-γH2AX (1:10,000). The HRP-conjugated secondary antibody for mouse or rabbit (Cell Signaling, 7074S for rabbit; RRID: AB_2099233, and 7076S for mouse) was diluted in 4% BSA/PBST by 1:10,000. After the membrane was soaked in ECL solution (BioRad), the blot signal was detected with a luminescent image analyzer (LAS4000, GE Healthcare).

Cells were deposited onto slide glasses (Superfrost Plus Microscope Slides precleaned, Fisher Scientific, 12–550–15) by cytospin. The deposited cells were pretreated with cold 0.1% Triton-X 100/PBS for 1 minute on ice (pre-extraction) and then fixed with 4% paraformaldehyde in PBS for 10 minutes. Then, cells were incubated with 5% BSA/PBST for 30 minutes (blocking step) and were incubated overnight with primary antibodies/5% BSA/PBST in a moisture chamber at 4°C by 1:300 dilution for phospho-S4/S8-RPA2, RPA2, and 1:1000 dilution for SLFN11. After washing with PBST, the cells were incubated with proper second antibodies/5% BSA/PBST by 1:1000 dilution for 2 to 4 hours. After washing with PBST, cells were mounted with mounting medium with DAPI (VECTOR, H-1200).

Signal intensity in each cell was measured using Fiji software. A fixed-sized circle that was slightly larger than a regular cell size was set. The same circle was used to measure the mean intensity of each signal throughout the experiments.

Cells were incubated with 10 μmol/L 5-bromo-2′-deoxyuridine (BrdU) or 5-ethynyl-2′-deoxyuridine (EdU) for 30 minutes before fixation with 70% ethanol or paraformaldehyde, respectively. BrdU was detected by anti-BrdU (Becton Dickinson, 347580) followed by Alexa Fluor 488 goat anti-mouse IgG (Molecular Probes, A110011). EdU was detected by click reaction by following the product manual (C10634, Thermo Fisher). Propidium iodide was used to measure DNA content. Data were collected with BD LSRFortessa (Becton Dickinson) or MACSQuant 10 (Miltenyi Biotec), and the data were analyzed with BD FACSDiva Software (RRID:SCR_001456) (Becton Dickinson) or MACSQuantify (Miltenyi Biotec).

Gene expression data, RNA-seq data, and drug activity data were obtained from NCI-60 (34), Genomics of Drug Sensitivity in Cancer (RRID:SCR_011956) (GDSC: https://www.cancerrxgene.org), and the Cancer Cell Line Encyclopedia (CCLE: https://portals.broadinstitute.org/ccle) using CellMinerCDB (https://discover.nci.nih.gov/cellminercdb/).

Statistical analyses were carried out using GraphPad Prism 7 software (GraphPad Prism, RRID:SCR_002798). For correlation analysis, Pearson's correlation was used and P < 0.01 was considered significant. For viability assays, a two-sided paired t test was used and P < 0.05 was considered significant.

The high-throughput drug screening data are available in the Supplementary Tables.

To identify drugs that exhibit cytotoxicity influenced by SLFN11, high-throughput drug screens were conducted at the NCATS using 1,978 oncology-focused compounds with mechanistic annotations (MIPE Library, v4; Supplementary Table S1 and S2; Fig. 1). We compared two isogenic pairs of human SLFN11-proficient and -deficient cell lines. The first pairwise comparison was between parental CCRF-CEM cells (an acute T-cell leukemia cell line with high endogenous SLFN11 expression) and CCRF-CEM SLFN11-knockout cells (Supplementary Table S1; Supplementary Fig. S1A). The second pairwise comparison was between an acute erythroleukemia cell line with deficient endogenous SLFN11 expression (K562 cells, K562 + empty vector) and K562 cells overexpressing WT SLFN11 (K562 + WT SLFN11; Supplementary Table S2; Supplementary Fig. S1A). Cellular viability was measured 48 hours after drug treatment with 11-point drug dose responses. Dose–response curves are classified into the CRCs, and half-maximal activity concentration (LAC₅₀) values were automatically estimated (see Materials and Methods; ref. 28).

To select hit compounds, first, we excluded compounds that exhibited a class 4 (inactive) or a class -3 (inconclusive hit) in SLFN11-positive cells (CCRF-CEM parent or K562 + WT SLFN11; ref. 28). Second, we excluded compounds exhibiting partial efficacy (50% or above viability at the highest concentration tested) in SLFN11-positive cells because of their low potency as clinical candidates. Third, we subtracted log₁₀[LAC₅₀ of SLFN11-negative cells] from log₁₀[LAC₅₀ of SLFN11-positive cells] for each drug and selected a compound in which differential log₁₀[LAC₅₀] is smaller than −0.1 (i.e., the drug is more potent in SLFN11-positive cells than the SLFN11-negative cells). Because LAC₅₀ cannot be obtained if the curve class is 4, compounds with valid LAC₅₀ in SLFN11-positive cells and not having LAC₅₀ in SLFN11-negative cells were also considered positive hits. Following this analysis, we retrieved 82 hit compounds in the CCRF-CEM isogenic pair (highlighted in yellow in Supplementary Table S1) and 125 hit compounds in the K562 isogenic pair (highlighted in yellow in Supplementary Table S2). The 29 overlapping drugs were selected as positive hits (Fig. 1A and Table 1).

As expected, the hit compounds included TOP1 and TOP2 inhibitors, antimetabolite pyrimidine antagonists, and replicative DNA polymerase inhibitors (Table 1; Supplementary Fig. S1). Although PARP inhibitors did not meet our hit selection requirements in the automated robotic screen, examination of the actual dose–response curves for olaparib and niraparib confirmed enhanced activity in SLFN11-positive cells (Supplementary Fig. S1) consistent with our previous reports (6). Another positive hit was PF-184, an inhibitor for the inhibitory factor-kappaB kinase 2 (IKK-2) that possesses anti-inflammatory activities. However, because 10 other IKK inhibitors included in the library did not show differential activity in our screening (Supplementary Tables S1 and S2), we concluded that “off-target” effects likely drove the PF-184 phenotypes. After a close examination of the screening data (Table 1), we decided to focus on two novel compounds whose activities had not previously been associated with SLFN11 biology: MLN-4924 (pevonedistat), an inhibitor of neddylation [targeting the NEDD8-activating enzyme (NAE)] under clinical development for cancer treatment (35), and AHPN, a retinoic acid receptor γ agonist (Fig. 1B). Other hit compounds were not further analyzed in this study because of their low potency as SLFN11-dependent drugs based on dose–response curves (Supplementary Fig. S2).

We next validated the high-throughput screening data results by repeating the cytotoxicity experiments and integrating an additional cell line, K562 overexpressing the putative-helicase dead SLFN11 (K562 + E669Q SLFN11) generated by introducing a point mutation in the Walker B motif of SLFN11 (20). We also tested TAS4464, another neddylation inhibitor with higher drug activity than MLN-4924 (31). Consistent with the results obtained with MLN-4924, we found that TAS4464 was more cytotoxic in the SLFN11-positive than in the SLFN11-negative cells (Fig. 1C). We verified that AHPN, also known as CD437, exhibited significant SLFN11-dependent drug sensitivity (Fig. 1C). Cytotoxicity of AHPN/CD437 is related to the induction of DNA damage and direct inhibition of DNA polymerase α (36), which causes replication stress. For MLN-4924, TAS4464, and AHPN/CD437, the drug sensitivity in K562 + E669Q SLFN11 was comparable to that in K562 + vector cells (Fig. 1C), indicating that the SLFN11-dependent sensitivity for these drugs is derived from the putative helicase activity of SLFN11. Together, these results identify two new classes of SLFN11-dependent drugs, neddylation inhibitors and the DNA polymerase α inhibitor AHPN/CD437.

As SLFN11 is recruited to chromatin within 4 hours in response to the replication stress induced by camptothecin (CPT) or the cell-cycle checkpoint 1 (CHEK1) inhibitor Prexasertib (20), we examined whether the two new hit compounds MLN-4924 and AHPN/CD437 also recruit SLFN11 to chromatin. We treated CCRF-CEM parental cells for 4 hours with near or above IC₉₀ of each drug. While AHPN/CD437 recruited SLFN11 to chromatin and induced SLFN11 foci at the nuclear periphery and inner nuclei like CPT or Prexasertib, MLN-4924 did not (Fig. 2A). As replication stress accompanies the phosphorylation of CHEK1, RPA2, and H2AX, we tested AHPN/CD437 and MLN-4924. Western blot analyses showed that AHPN/CD437 induced phosphorylation of these proteins within 4 hours, but MLN-4924 did not (Fig. 2B). Moreover, cell-cycle analyses revealed that AHPN/CD437 strongly blocked replication while MLN-4924 did not after 4-hour treatments (Fig. 2C). These results confirm that AHPN/CD437 but not MLN-4924 behaves as a DNA damaging agent directly targeting ongoing DNA replication (36).

To elucidate how MLN-4924 engages SLFN11, we performed time-course experiments beyond the 4-hour time point. The target of MLN-4924 is NAE, an essential E1 enzyme of the NEDD8 conjugation (neddylation) pathway (37). Neddylation is required to activate cullin-RING ubiquitin ligase complexes (CRL). Hence, neddylation inhibition by MLN-4924 leads to the accumulation of CRL substrates through the inhibition of their proteasome-dependent degradation (37, 38). Among the CRL substrates stabilized by MLN-4924, CDT1 (chromatin licensing and DNA Replication factor 1), an essential factor of origin firing, is a major effector for the cytotoxic effects of MLN-4924 (39, 40). Because CDT1 needs to be degraded once used for origin firing, accumulation of CDT1 leads to aberrant re-replication (35, 40).

CDT1 accumulation was observed at 4 hours and maintained until 12 hours after MLN-4924 treatment regardless of SLFN11, while phosphorylated RPA2 remained unchanged (Fig. 3A). Phosphorylation of RPA2 became obvious 24 hours after MLN-4924 treatment with continuous CDT1 stabilization regardless of SLFN11. These observations imply that replication stress was induced by MLN-4924 in both SLFN11-expressing and non-expressing cells (Fig. 3B). Consistently, MLN-4924 treatment increased the chromatin loading of RPA2, a hallmark of replication stress regardless of SLFN11 (Fig. 3C; Supplementary Fig. S3A). Moreover, in the SLFN11-expressing cells, SLFN11 was recruited to chromatin with foci formation at the nuclear periphery and inner nucleus, where phospho-RPA2 was partially colocalized (Fig. 3D; Supplementary Fig. S3B). These results reveal that it takes 24 hours for MLN-4924 to exert a sufficient level of replication stress for SLFN11 recruitment to chromatin. This time course is much longer than those by previously reported DNA-targeting drugs showing SLFN11-dependent toxicity.

As SLFN11 blocks abnormal (stressed) replication forks (20), we analyzed the cell cycle of SLFN11-positive or -negative cells treated with MLN-4924. At 24- and 48-hour treatments, the population of nonreplicating S-phase cells (2N<, <4N shown in yellow-green) increased more in SLFN11-positive cells than in SLFN11-negative cells (Fig. 3E). These nonreplicating cells do not appear apoptotic as judged by the absence of γH2AX pan-staining (ref. 41; Supplementary Fig. S3C). By contrast, re-replicating cells (>4N and BrdU-positive cells shown in red in Fig. 3E) increased more in the SLFN11-KO cells than in the parent cells at 48 hours. A high dose (1,000 nmol/L) of MNL-4924 induced more re-replicating cells in the SLFN11-KO cells (Supplementary Fig. S4A). Consistent results were obtained in comparing K562 + WT SLFN11 and K562 + E669Q SLFN11 (Fig. 3F), indicating that the re-replication block by SLFN11 depends on the putative helicase motif of SLFN11. Using another isogenic pair of prostate cancer DU145 cells with high endogenous SLFN11 (6), we confirmed that MLN-4924 induced more re-replication in SLFN11-KO cells than in the parental cells (Supplementary Fig. S4B). On the basis of these results, we conclude that SLFN11 blocks unscheduled re-replication induced by CDT1 accumulation in response to Pevondistat.

To expand our findings of SLFN11-dependent sensitization to neddylation inhibitors to non-isogenic cell lines, we mined the data of three independent cancer cell databases [NCI-60, Cancer Therapeutics Response Portal (CTRP) and Genomic of Drug Sensitivity in Cancer (GDSC)] for the correlation between SLFN11 gene expression and the sensitivity to MLN-4924 (Fig. 4A; https://discover.nci.nih.gov/; refs. 34, 42). The correlations were significantly positive in the three datasets, consolidating our isogenic cell line data. We also expanded the findings to the other neddylation inhibitor, TAS4464, as its IC₅₀ has been determined across 235 cell lines from various tissues (31). SLFN11 expression level was available for 179 of the 235 cell lines from the CTRP database (Supplementary Table S3). SLFN11 expression and sensitivity to TAS4464 were also significantly correlated (Fig. 4B). Overall, we conclude that both neddylation inhibitors, MLN-4924 and TAS4464 are more active in SLFN11-positive cells than SLFN11-negative cells in a broad range of cell types, and that SLFN11 expression might be a potential biomarker for drug response in clinical trials with neddylation inhibitors.

In addition to SLFN11 acting as a potent determinant of response to neddylation inhibitors, we explored pathways necessary for repairing neddylation inhibitor-induced DNA damage. We employed a preestablished DNA-repair mutant library of chicken DT40 cell lines (29), including WT and mutants of nonhomologous end-joining (Ku70 or Ligase IV), homologous recombination (Brca2 or Mre11), and translesion synthesis (Pcna, Polz or Polh), Parp1, nuclease excision repair (Xpa or Xpb), and Fanconi anemia (Fancd2 or Fancg) genes. Note that chicken cells do not have the SLFN11 gene (43). We measured the IC₉₀ values of cisplatin, MLN-4924, and TAS4464 across 13 cell lines and calculated the relative sensitivity of each mutant compared to WT DT40 cells (Fig. 5). The data revealed that, as reported, homologous recombination, translesion synthesis, and Fanconi anemia genes are specifically required to repair cisplatin-induced lesions. On the other hand, MLN-4924 and TAS4464 induced promiscuous DNA lesions that required various DNA repair pathways including nonhomologous end-joining, which is different from cisplatin (Fig. 5). These results indicate that MLN-4924 and TAS4464 can be potent in a broad range of DNA-repair–deficient cancer cells. However, at the same time, it will be hard to determine any specific biomarkers in DNA repair genes.

To explore other biomarkers for MLN-4924, we mined the resource of the sensitivity profile to MLN-4924 examined in a gene-disruption mutant library generated in human RPE cells (44). Because RPE cells express SLFN11 at a low level, SLFN11-dependent sensitivity to MLN-4924 was not validated with this database. We picked the top ∼150 genes that exhibit heightened sensitivity to MLN-4924 when disrupted and analyzed them for pathway analysis using the Reactome (https://reactome.org). The most significant pathway was Cell Cycle followed by Antigen Processing and Homologous Recombination. Hence, further studies are warranted to explore whether specific cell-cycle genes may represent other potential candidate biomarkers of neddylation inhibitors beyond SLFN11.

SLFN11 has been an emerging target for cancer therapy since the first discovery of its relevance to drug sensitivity in 2012 (2, 3). Until now, the SLFN11-dependent drugs were known to recruit SLFN11 to chromatin within 2 to 4 hours. Here, we report two additional SLFN11-dependent drug classes. One is AHPN/CD437, which behaves similarly to the previously reported SLFN11-dependent drug (refs. 4, 18, 19; with rapid induction of SLFN11 recruitment to chromatin), and the other is neddylation inhibitors [MLN-4924 (pevonedistat) and TAS4464] that activate SLFN11 by unscheduled re-replication mediated by supraphysiologic accumulation of CDT1 (40, 45). We also provide the first evidence that SLFN11 uniquely blocks unscheduled re-replication induced by MLN-4924. Because neddylation inhibitors are in clinical development, our results demonstrate the prominence of replication stress as a cytotoxic anticancer mechanism of pevonedistat (MLN-4924) and TAS4464. They also suggest the inclusion of SLFN11 as a predictive biomarker for neddylation inhibitors in clinical trials.

We show that SLFN11-positive cells are more sensitive to MLN-4924 or TAS4464 than SLFN11-negative cells in analyses of isogenic cell lines (Fig. 1) and in a broad range of cancer cell lines from different tissues of origin (Fig. 4). Yet, the significance is less prominent than those for other DNA-targeting agents such as TOP1 and TOP2 inhibitors (Supplementary Fig. S1). For example, the P value between SLFN11 expression and MLN-4924 activity is 0.012, while it is 3.6e-13 between SLFN11 expression and the TOP1 inhibitor camptothecin in the NCI-60 database (https://discover.nci.nih.gov/). This is likely related to the fact that the anticancer activity of neddylation inhibitors is not merely derived solely from replication stress but also from the alterations of multiple cellular pathways. For instance, MLN-4924 causes the accumulation of proapoptotic proteins (NOXA, BIK, and BIM) and the downregulation of antiapoptotic proteins (BCLxL) to trigger apoptosis (37). Inhibition of the NF-kB pathway is another major cause of MLN4924-induced cytotoxicity, as evidenced in acute myeloid leukemia (46). Hence, MLN-4924 can kill cells by additional mechanisms from the SLFN11-mediated replication block.

Neddylation inhibition is more effective in cycling cells than in non-cycling cells (39), indicating that re-replication due to CDT1 stabilization is a major contributor to the cytotoxicity of MLN-4924. This is evidenced by the fact that CDT1 knockdown in cancer cells partially reverses the cytotoxic effect of MLN-4924 (39). Re-replication is distinct from endoreplication, which is the developmentally controlled endo-cycle with discrete periods of S-phase and G-phase (25). Re-replication causes extensive single-and double-strand DNA damage through the collision of replication forks, which generates RPA foci that activate ATR, a hallmark of replication stress at a later point (24 hours; ref. 45; Fig. 3). We also show that SLFN11 is recruited to chromatin 24 hours after MLN-4924 treatment when SLFN11 blocks unscheduled re-replication (Fig. 3). Therefore, both the re-replication in the absence of SLFN11 and the re-replication block by SLFN11 contribute to the cytotoxicity of MLN-4924. Yet, our study reveals that the latter is more cytotoxic, suggesting that some cells that undergo re-replication can survive.

Although we reported that SLFN11 degrades CDT1 via a ubiquitin-proteosome pathway under replication stress (23), the CDT1 stabilization by neddylation inhibition occurs upstream of the ubiquitination (i.e., ubiquitination cannot take place if neddylation is inhibited). Hence, SLFN11-mediated CDT1 degradation is not apparent in our experiments using MLN-4924 (Fig. 3A and B).

Clinical trials of MLN-4924 (pevonedistat) have shown that the compound is well tolerated (38). According to the latest review on neddylation inhibitors, MLN-4924 has been employed in more than 40 clinical trials under phase I/II/III alone or in combinations with chemo-/radiotherapy (https://www.clinicaltrials.gov/; ref. 47). The trials include acute myelogenous leukemia, myelodysplastic syndrome, lymphoma, melanoma, solid tumors, and others, without predictive biomarkers. Because SLFN11 expression level varies across tumor types (∼0% in colon cancer and ∼100% in Ewing sarcoma) and within tumor types (2, 48), measuring SLFN11 by IHC and by its transcript levels can be justified to determine whether SLFN11 expression predicts response to pevonedistat and other neddylation inhibitors in clinical trials (19, 48).

Clinical development of TAS4464 seems currently discontinued (49). Yet, TAS4464 can potentially treat not only hematologic malignancies but also solid tumors in a preclinical model across 240 human tumor cell lines of various tissue of origin and molecular backgrounds (31). Although single biomarkers that can predict the TAS4464 sensitivity were not identified in the analysis of multiple cell lines (31), our reexamination of their data combining SLFN11 mRNA expression levels obtained from CTRP databases identified a significant correlation between SLFN11 expression and the sensitivity to TAS4464 (Fig. 4B). Overall, our study with the neddylation inhibitors demonstrates that SLFN11 senses and blocks unscheduled re-replication, and that its use as a clinical biomarker could be extended to neddylation inhibitors.

J. Murai reports nonfinancial support from Taiho Pharmaceutical Co., Ltd., grants from Japan Grants-in-Aid for Scientific Research and the Japan Agency for Medical Research and Development during the conduct of the study, as well as grants from Taiho Pharmaceutical Co., Ltd. outside the submitted work. No disclosures were reported by the other authors.

J. Murai: Conceptualization, formal analysis, funding acquisition, validation, methodology, writing–original draft. M. Ceribelli: Data curation, formal analysis. H. Fu: Data curation, formal analysis. C.E. Redon: Data curation, formal analysis. U. Jo: Data curation, methodology. Y. Murai: Data curation, methodology. M.I. Aladjem: Resources. C.J. Thomas: Resources, software, methodology, project administration. Y. Pommier: Conceptualization, resources, supervision, funding acquisition, investigation, methodology, writing–original draft.

This project was supported by the Intramural Program, Center for Cancer Research of the NCI, NIH (Z01 BC 006161 and 006150) (to Y. Pommier), AMED (Japan Agency for Medical Research and Development), Project for Cancer Research and Therapeutic Evolution (to J. Murai), and Grants-in-Aid for Scientific Research (JP19H03505; to J. Murai) from the Japan Society for the Promotion of Science. This work was also supported by research funds from the Yamagata prefectural government and the City of Tsuruoka (to J. Murai).

We would like to thank the staff at the NCATS who contributed to this study.

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.