Destabilization of the human RED–SMU1 splicing complex as a basis for host-directed antiinfluenza strategy

Significance

Influenza virus is a serious threat to global public health, and there is a critical need for innovative antiinfluenza drugs. Two broad nonexclusive approaches to inhibit viral replication are possible, either targeting viral proteins directly or targeting host proteins essential for the viral life cycle. Here, we took the second approach, which is more likely to counter the problem of drug-resistant virus emergence. Focusing on an essential host partner of influenza viruses, the RED–SMU1 splicing complex, we performed a virtual structure-based drug screening. We identified two synthetic molecules that interfere with RED–SMU1 complex assembly, inhibit the splicing of viral messenger RNAs, and show potential for the inhibition of influenza virus infections.

Abstract

New therapeutic strategies targeting influenza are actively sought due to limitations in current drugs available. Host-directed therapy is an emerging concept to target host functions involved in pathogen life cycles and/or pathogenesis, rather than pathogen components themselves. From this perspective, we focused on an essential host partner of influenza viruses, the RED–SMU1 splicing complex. Here, we identified two synthetic molecules targeting an α-helix/groove interface essential for RED–SMU1 complex assembly. We solved the structure of the SMU1 N-terminal domain in complex with RED or bound to one of the molecules identified to disrupt this complex. We show that these compounds inhibiting RED–SMU1 interaction also decrease endogenous RED-SMU1 levels and inhibit viral mRNA splicing and viral multiplication, while preserving cell viability. Overall, our data demonstrate the potential of RED-SMU1 destabilizing molecules as an antiviral therapy that could be active against a wide range of influenza viruses and be less prone to drug resistance.

The increasing incidence of drug-resistant pathogens calls for the development of novel therapeutic strategies. In recent years, the concept of host-directed therapies, which target host determinants essential for the infectious life cycle and/or pathogenesis rather than pathogen components, has been rapidly expanding (1, 2). Preclinical studies suggest they could show clinical safety while providing the advantage of broad-spectrum efficacy and reduced antiviral resistance. A range of host-directed therapies for several bacterial and viral diseases, with different mechanisms of action, are currently in clinical trials (1, 2).

Influenza A viruses (IAVs) are a leading cause of morbidity and mortality worldwide, as they are responsible for recurring annual epidemics, frequent epizootics, and occasional pandemics (3). The drugs currently licensed for treatment of influenza target viral components (the M2 ion channel, the neuraminidase, and the polymerase), and they all lead to the emergence of resistance variants. The M2 inhibitors are no longer recommended for use since currently circulating human H3N2 and H1N1pdm09 viruses have become naturally resistant to these drugs (4). Although the frequency of resistance to neuraminidase inhibitors (NAIs) of currently circulating human IAVs remains low (around 0.5%) (5), large clusters of H1N1pmd09 viruses resistant to oseltamivir, the most widely used NAI, have been observed (6, 7). The former seasonal H1N1 IAVs had become globally oseltamivir-resistant during the 2007–2008 season (8). Favipiravir, a purine nucleoside analog also known as T-705, is undergoing phase III trials in America and Europe and has been approved in Japan to treat pandemic influenza virus infections. In October 2018, the US Food and Drug Administration (FDA) approved Xofluza, a selective inhibitor of the polymerase PA subunit, for the treatment of acute uncomplicated influenza. However, the first evidence for viral adaptation to favipiravir treatment in cell culture has been reported recently (9, 10), and the emergence of PA variants with a mutation conferring resistance to Xofluza was observed in 9.7% of treated patients in the phase III trial (11). In this context, novel antiinfluenza drugs are actively being sought to efficiently fight IAVs, particularly with regard to pandemic preparedness (12, 13). Antivirals currently in late-phase clinical trials include monoclonal antibodies against the viral hemagglutinin (HA) or matrix protein (14, 15), a selective inhibitor of the polymerase PB2 subunit (16), and two compounds targeting host cell factors: nitaxozanide, which impairs trafficking and maturation of the viral HA (17, 18), and DAS181, which enzymatically removes the membrane receptors required for IAV attachment to target cells (19, 20). Continuous progress on the identification of host factors involved in the IAV life cycle provides a basis for the development of alternative host-directed antiviral drugs (21, 22). Following this approach, we have characterized the structure of a human splicing factor, RED-SMU1, which is essential for the IAV life cycle, and investigated this cellular factor as a potential target for IAV therapy.

Infectious IAV particles contain eight viral ribonucleoprotein complexes, corresponding to a set of eight distinct viral genomic RNA segments encapsidated with nucleoproteins and associated with the heterotrimeric viral RNA-dependent RNA influenza polymerase (FluPol) consisting of the PB1, PB2, and PA subunits (3). The viral genome has a limited size (about 13.5 kb), but IAVs have evolved a variety of strategies to expand their coding capacity. Viral mRNAs are synthesized by the FluPol in the nucleus of infected cells. Although most of these are intronless, the M, NS, and PB2 segments produce both unspliced mRNAs (M1, NS1, and PB2) and spliced mRNAs (M2, NS2, and PB2-S1). We previously showed that FluPol recruits a complex formed by the human splicing factors RED (78.9 kDa) and SMU1 (57.5 kDa) by direct binding to RED (23). RED-SMU1 regulates the splicing of viral NS1 mRNA into the mRNA encoding the multifunctional and essential NS2/NEP protein. In cells depleted for RED or SMU1, the production of infectious influenza virions showed a 100-fold reduction (23), designating the RED–SMU1 complex as a promising IAV drug target.

The RED and SMU1 factors are associated with the precatalytic spliceosome (24). They were found to jointly regulate the splicing of specific cellular pre-mRNAs, with a possible role in the selection of splice sites (25–27). As we and others have observed that RED and SMU1 directly bind (23, 26, 28) and stabilize each other (23, 26, 27), a significant proportion of functional RED–SMU1 complexes is likely to exist.

Here, we performed subdomain analysis of human RED and SMU1 proteins using cell-based interaction assays, and solved the crystal structure of a minimal RED–SMU1 complex at 3.0-Å resolution. The structure shows similarities to the recently published RED and SMU1 orthologs in Caenorhabditis elegans (ceRED and ceSMU1) (29) but reveals a more intricate RED–SMU1 interface. Based on our structural findings, we identified small molecules that target an α-helix/groove interface essential for RED–SMU1 interaction. Lastly, we provide supporting data to demonstrate the potential of such RED-SMU1 destabilizing molecules for the inhibition of IAV infections.

Results

Delineation of a Minimal RED–SMU1 Complex.

We previously found that the N-terminal domain of RED (residues 1–315) binds to SMU1 as efficiently as the full-length RED protein (23). Using a split-luciferase–based interaction assay, we further delineated a minimal human RED–SMU1 complex. Plasmids encoding various subdomains of RED tagged with the Gluc2 fragment of Gaussia princeps luciferase were cotransfected with plasmids for the expression of SMU1 or SMU1 N-terminal region (SMU1_Nter; residues 1–196) fused to the transcomplementing G. princeps luciferase Gluc1 fragment. The rationale for RED truncations shown in Fig. 1A was based on secondary structure and disorder predictions (SI Appendix, Fig. S1), with an iterative testing and design approach. As revealed by the relative luciferase activities measured in cell extracts, the RED[206–260] subdomain [RED central region (RED_mid)] retained the interaction with SMU1 (Fig. 1A, black bars) or SMU1_Nter (Fig. 1A, gray bars). The SMU1 interaction signals were almost eightfold higher with RED_mid than with the full-length RED (P < 0.01), and were higher than with any longer truncated version of RED. Western blot analysis of the cell lysates used for luciferase assay showed that this higher interaction signal was not due to a higher level of expression of RED_mid (Fig. 1B, black arrowhead) compared with other deletion mutants of RED or with the full-length RED (Fig. 1B, open arrowhead). A likely explanation is that the short RED_mid subdomain is more stable thermodynamically than the longer versions of RED, and that it becomes more accessible upon removal of the adjacent polypeptides. For several other protein pairs, it was indeed previously shown that interactions are more easily detected when using isolated interacting subdomains or peptides (30). Notably, the RED[219–299] subdomain produced lower interaction signals compared with RED_mid, but also compared with the full-length RED protein, suggesting that residues 206–218 of RED are essential for the interaction with SMU1.

Fig. 1.

RED–SMU1 interaction mapping using a cell-based assay. The G. princeps luciferase-based complementation assay was performed as described in Materials and Methods. Gluc2-fused RED full-length (RED) or indicated subdomains were coexpressed with Gluc1-fused SMU1 full-length (SMU1) or SMU1[1–196] (SMU1_Nter) by transient transfection in HEK-293T cells. (A) Normalized luciferase activities are expressed as percentages relative to the activity measured with full-length RED and SMU1 proteins. The data shown are the mean ± SD of three independent experiments in triplicate. Black and gray bars represent luciferase activities measured in the presence of SMU1 and SMU1_Nter, respectively. **P < 0.01 (parametric unpaired t test). On the schematic diagram of RED, its characteristic stretch of repeated arginine, glutamic acid, and aspartic acid residues is represented by a hatched box. The dotted lines highlight the interaction between the two components of the minimal RED_mid–SMU1_Nter complex. (B) Cell lysates used to determine luciferase activities in A were subsequently analyzed by Western blot using antibodies specific for G. princeps luciferase (Gluc, Upper) and glyceraldehyde phosphate dehydrogenase (GAPDH, Lower). The white arrowhead points to Gluc1-SMU1; the open arrowhead points to Gluc2-RED; and the black arrowhead points to Gluc2-RED[206–260], renamed Gluc2-RED_mid in the text.

The 3D Structure of the Human RED_mid–SMU1_Nter Complex Reveals Two Binding Interfaces.

Based on our cell-based interaction results, we set up a coexpression strategy to produce and pull down the human RED_mid–SMU1_Nter complex, using a hexahistidine tag fused to the N terminus of RED_mid. The recombinant complex was crystallized, and the structure was solved by molecular replacement using the X-ray structure of the SMU1_Nter alone without RED_mid obtained earlier (SI Appendix, Fig. S2 and Table S1). The final model was built using the anomalous signal of an Se-methionine (Met) derivative dataset for an accurate sequence assignment. The asymmetrical unit is composed of 16 molecules: eight SMU1_Nter molecules that assemble into four dimers and eight RED_mid molecules, each one associated with a SMU1_Nter monomer. The X-ray structure reveals two distinct interfaces between SMU1_Nter and RED_mid (Fig. 2A). First, the region of RED_mid corresponding to residues 235–257 disrupts the initial fold of the N terminus of SMU1_Nter to form a three-stranded antiparallel β-sheet that is only stabilized by main chain interactions (interface I, Fig. 2B). Second, the helical region of RED_mid (corresponding to residues 211–221) lies in a hydrophobic groove delimited by three continuous α-helices (α4 to α6) on the surface of SMU1_Nter (interface II, Fig. 2C). The structural results are consistent with the high- and low-interaction signals measured with the SMU1-RED[206–260] and SMU1-RED[219–299] combinations, respectively (Fig. 1A).

Fig. 2.

Crystal structure of the recombinant human RED_mid–SMU1_Nter complex. (A) Global view of the complex. The RED_mid and SMU1_Nter proteins are colored red and yellow, respectively. One monomer of SMU1_Nter is shown as a ribbon diagram and the second as a surface representation. (B) Close-up view of the mixed β-sheet formed by the interaction of RED[235–257] with the N-terminal β-strand of a SMU1_Nter monomer and stabilized by the C-terminal surface of the other dimer partner SMU1_Nter molecule. (C) Details of the second interface corresponding to the α-helical part of RED_mid inserted into a hydrophobic groove at the surface of SMU1_Nter. The images were generated using PyMOL (77). The atomic coordinates are in PDB entry 6Q8I.

The comparison of our human RED_mid–SMU1_Nter complex structure with the recently published structure of a Caenorhabditis elegans minimal RED–SMU1 complex (29) (SI Appendix, Fig. S3) is discussed below. Importantly, the human structure suggests that SMU1 needs to form a dimer to assemble with RED, as interface I extends across two distinct SMU1_Nter monomers. In agreement with this observation, a size-exclusion chromatography combined with multiangle laser light scattering (SEC-MALLS) analysis on purified full-length SMU1 protein showed that it is mostly dimeric (SI Appendix, Fig. S4A), and treatment of A549 or HEK-293T cells with the cell-permeable cross-linking agent disuccinimidyl suberate followed by Western blotting revealed that the endogenous SMU1 protein forms dimers in live cells (SI Appendix, Fig. S4B).

Mutational Analysis Provides a Rationale for Targeting the α-Helix/Groove Interface.

To confirm our structural data and assess the druggability of the RED–SMU1 interface, we performed structure-based mutational analysis of the full-length RED and SMU1 proteins. Point mutations were introduced at interface I (Fig. 2B) or II (Figs. 2C and 3 A and B). Split-luciferase complementation assays were performed, and expression levels of the mutants were assessed (Fig. 3 C–H). Short deletions in the domains of SMU1 (Δ3–7) or RED (Δ237–241 or Δ254–256) involved in interface I reduced the RED–SMU1 interaction signal up to fivefold (Fig. 3C, black bars). In the presence of SMU1 mutations targeting interface II (D57Q, D57A-E89A, L60D-I63D, L73D-Y77D, and L84D-L87D) or the hinge between interfaces I and II (P157T-P158T), a two- to 10-fold reduction of the interaction signal was also observed (Fig. 3D). Notably, no cumulative effect was observed when deletions at interface I were combined with the SMU1 mutations D57A-E89A destabilizing interface II (Fig. 3C, gray bars), suggesting that both molecular interfaces are required to stabilize the RED–SMU1 complex. RED mutants L212D, V216D, and L220D targeting interface II showed a twofold reduction of the interaction signals with SMU1 or SMU1-D57Q (Fig. 3E). In contrast the RED-N215D and M219D mutations had little effect on the interaction with SMU1 (Fig. 3E), as expected from the 3D structure (Fig. 3 A and B). Together, these results confirm our structural findings; demonstrate that a mere disruption of interface II α-helix/groove impairs the stability of the RED–SMU1 complex; and point to SMU1 D57, L60, I63, L73, Y77, L84, L87, and E89 as key RED-binding residues.

Fig. 3.

Structure-based analysis of the RED–SMU1 interface. (A) Schematic representation of interface II. On the helical wheel representation of RED (residues 213–222), aliphatic residues (in a gray frame) are concentrated at the RED–SMU1 interface. The polar residues (in blue) are exclusively located on the opposite side. (B) Location of SMU1 and RED residues (colored green and purple, respectively) involved in interactions at interface II and subjected to the mutagenesis experiments shown in C–H. (C–E) Luciferase-based complementation assays were performed as described in Materials and Methods with the indicated combinations of wild-type (wt) or mutant proteins. The normalized luciferase activities are expressed as percentages relative to the activity measured in the presence of the wt SMU1 and RED proteins. The data shown are the mean ± SD of three independent experiments in triplicate, except when SMU1-D57A-E89A was tested in C (two independent experiments). **P < 0.01, ***P < 0.001 (parametric unpaired t test). (F) Indicated combinations of wt or mutant SMU1 and RED proteins, fused to the Strep-tag, were transiently coexpressed in HEK-293T cells. (Lower) Cell lysates were analyzed by Western blot, using HRP-conjugated Strep-Tactin. (Upper) GAPDH was used as a loading control. (G and H) Cell lysates used to determine luciferase activities in D and E were subsequently analyzed by Western blot, using antibodies specific for Gluc (Lower) and an anti-GAPDH antibody (Upper).

Disrupting the RED–SMU1 Complex: Proof of Principle by Overexpressing RED_mid.

Based on the data presented above, we reasoned that if overexpressed, the RED_mid helical domain should efficiently compete with the full-length RED protein for binding to the surface groove on SMU1, and thereby disrupt the RED–SMU1 complex. As a control, we used a RED_mid-V216D mutant, which accumulates at the same levels as RED_mid but is strongly impaired for SMU1 binding. Of note, the effect of V216D mutation on SMU1 binding is much stronger in the context of RED_mid (>99% reduction) compared with the full-length RED (about 50% reduction) (SI Appendix, Fig. S5A). Upon overexpression of the mCherry-RED_mid fusion protein, we observed a greater than twofold reduction of the RED–SMU1 interaction signal, as measured with the split-luciferase complementation assay, while with the control mCherry-RED_mid-V216D mutant, no such reduction was observed (P < 0.01; SI Appendix, Fig. S5 B and C).

We then assessed the effect of RED_mid overexpression on the endogenous RED–SMU1 complex. As there is evidence that RED and SMU1 stabilize each other (23, 26, 27), RED-SMU1 disruption is expected to result in decreased steady-state levels of the proteins. We observed a decrease in endogenous cellular levels of RED when overexpressing RED_mid, whereas empty vectors or RED_mid-V216D had no effect (SI Appendix, Fig. S5D, Upper). However, SMU1 levels remained unchanged (SI Appendix, Fig. S5D, Middle). A likely interpretation is that the SMU1 protein complexed with RED_mid remains stable, whereas the dissociated RED protein undergoes degradation.

Next, we monitored the effect of RED_mid overexpression on the replication of a recombinant A/WSN/33 virus carrying a luciferase reporter gene (WSN-PB2-Nanoluc). The luciferase activity measured in cell lysates prepared at 24 h postinfection (hpi), was lower in cells overexpressing RED_mid compared with cells transfected with the empty pCI vector or overexpressing RED_mid-V216D (SI Appendix, Fig. S5E). On the contrary, there was no significant difference between cells transfected with the empty pCI vector and those overexpressing RED_mid-V216D. Furthermore, splicing of the viral NS1 mRNA into NS2 mRNA, which is dependent upon RED-SMU1 (23), was reduced in RED_mid-expressing cells compared with the RED_mid-V216D control (SI Appendix, Fig. S5F). Overall, our findings strengthen the rationale of targeting the RED-SMU1 α-helix/groove interface to inhibit IAV replication.

In Silico Identification and Evaluation of RED-SMU1 Disrupting Compounds.

We used the available structural information on the RED_mid–SMU1_Nter complex to perform molecular docking-based screening for compounds targeting the RED-binding groove at the surface of SMU1_Nter. The virtual high-throughput screening (vHTS) was performed on a set of 4,121 chemical compounds from our in-house chemical database [Laboratoire de Chimie et Biochimie Pharmacologiques et Toxicologiques-DataBase (LCPBT-DB)], comparable to the Prestwick- and FDA-approved libraries in terms of chemical diversity (SI Appendix). A flowchart for the vHTS and compound selection pipeline is shown in Fig. 4A. To ensure a time/precision ratio compatible with vHTS, we used a protocol in which amino acid side chains of the protein are left flexible only around the binding site, and we tested 10 random conformers for each ligand. The resulting poses were ranked according to their docking scores, as described in Materials and Methods. The poses with the highest docking scores were further filtered to select for compounds that are predicted to bind at least four of the eight SMU1 residues required for efficient binding to RED: D57, L60, I63, L73, Y77, L84, L87, and E89 (Figs. 3D and 4B). A total of 37 compounds fulfilled both criteria, and a subset of 16 molecules with representative chemical scaffolds were further evaluated.

Fig. 4.

SMU1_Nter binding to compounds LSP641 and LSP61. (A) Flowchart of in silico screening and compound evaluation. The indicated assays were performed as described in Materials and Methods, in the presence of compounds at 60 μM or DMSO. The Venn diagrams show the number of selected compounds. The cutoffs applied are indicated in italics (cell viability assay: less than twofold reduction in CellTiter-Glo signal after 36-h incubation with the compound compared with DMSO; RED-SMU1 and IAV replication assays: ≥20% and ≥80% reduction in Gaussia luciferase and Nanoluc signal, respectively, in the presence of the compound compared with DMSO). (B) SMU1_Nter residues required for efficient binding to RED (colored green) and used for filtering of the docking poses. (C) Chemical structures of compounds LSP641 and LSP61. (D) Cocrystal structure of LSP641 in complex with SMU1_Nter. The LSP641-binding pocket is colored according to hydrophobicity (blue, hydrophilic; red, hydrophobic). The atomic coordinates are in PDB entry 6Q8J. (E) Residues of SMU1 involved in LSP641 binding. Key residues for RED binding that also interact with LSP641 (green) and other residues that strongly interact with LSP641 in the cocrystal structure (orange) are shown. (F) Representative pose of LSP61 upon in silico docking on SMU_Nter. The LSP61-binding pocket is colored according to hydrophobicity (blue, hydrophilic; red, hydrophobic). (G) Residues of SMU1 predictively involved in LSP61 binding. Key residues for RED binding that also interact with LSP61 upon molecular docking (green) and other residues that strongly interact with LSP61 upon molecular docking (orange) are shown.

Among these, 14 showed no or only limited cytotoxicity at 60 μM and were further tested at the same concentration in cell-based assays for the inhibition of RED–SMU1 interaction and IAV replication. Compounds inhibiting RED–SMU1 interaction by more than 20% and IAV replication by more than 80% were selected for further evaluation. In total, three compounds showed promising inhibitory effect in both assays (Fig. 4A). Of these, one molecule, LSP641 (Fig. 4C), was successfully cocrystallized with the purified recombinant SMU1_Nter domain. Structural data showed that LSP641 locates in the hydrophobic RED-binding pocket of SMU1 as expected (Fig. 4D) and is engaged in several molecular interactions with SMU1 (Fig. 4E). The 2-amino-pyrimidine group present in LSP641 forms key hydrogen bounds with Q61 and Q64, whereas the urea group forms a hydrogen bond with Y77. In the apolar region of SMU1, D57, L60, and L96 are involved in Pi–Pi or σ–Pi interactions, particularly with the pyridopyrimidine scaffold of LSP641. Furthermore, W56, I63, L84, and A92 form hydrophobic interactions with LSP641. Remarkably, the LSP641 molecule interacts with D57, L60, I63, Y77, and L84 (i.e., five of the eight residues shown earlier to be involved in RED–SMU1 interaction) to account for the inhibition of this protein–protein interaction.

Guided by the SMU1_Nter–LSP641 costructure, we selected 27 LSP641-related compounds and we filtered them through a second evaluation round (Fig. 4A), which led to the selection of LSP61 (31) (Fig. 4 C and F). Unfortunately, we were unable to cocrystallize LSP61 with SMU1_Nter. This different behavior of LSP61 compared with LSP641 might be related to the absence/presence of stacking interactions in the cocrystal. However, in silico molecular docking supported binding interactions similar to LSP641 (Fig. 4G). Hydrogen bonds with Q64 and Y77 are conserved, as well as Pi–Pi or σ–Pi interactions with D57 and L60, and hydrophobic interactions with W56 and L84. Q61 is involved in Pi interactions, whereas V41, V80, and L96 are forming novel hydrophobic interactions. Most importantly, the benzylpiperidine group of LSP641 that is not binding SMU1 in the crystal structure (Fig. 4 D and E) is replaced in LSP61 by an alkyl chain that interacts with A62 and V41 in a neighboring hydrophobic pocket of SMU1, and therefore optimizes molecular recognition (Fig. 4 F and G). This led us to select LSP61 to conduct further investigations, in parallel with LSP641.

LSP61 and LSP641 were compared in the RED–SMU1 interaction assay across the concentration range of 0.23–60 µM. The EC₅₀ (half-maximal effective concentration) was estimated to be 15 μM for LSP61; LSP641 showed a milder effect resulting in 40% inhibition at 60 μM, compared with 90% with LSP61, and a plateauing curve (Fig. 5A). When a similar FOS/JUN interaction assay was used as a specificity control, neither of the two compounds showed a significant effect (Fig. 5B). The K_d of the SMU1_Nter–LSP61 complex could not be determined, most likely due to the low solubility of LSP61 (logP = 4.8). The effect of LSP61 on the endogenous RED–SMU1 complex was assessed using the steady-state levels of RED and SMU1 as a proxy. Expression levels of both proteins decreased when treating cells with 15, 30, and 60 μM LSP61 (Fig. 5C). At these concentrations, no cytotoxicity, but some cytostatic effect, was observed as assessed by the steady levels of ATP measured in culture wells (a proxy for cell count) upon 24 and 48 h of incubation with LSP61 (Fig. 5D) and the observation of cell monolayers by bright-field microscopy (SI Appendix, Fig. S6).

Fig. 5.

Evaluation of compound LSP61 for RED-SMU1 destabilization. (A) RED–SMU1 interaction assay. The split-luciferase assay was performed as described in Materials and Methods. At 8 h posttransfection (hpt), the A549 cells were incubated with the relevant compounds at the concentrations indicated. The normalized luciferase units measured at 24 hpt are expressed as percentages relative to the DMSO-treated control. The data shown are the mean ± SD of three independent experiments in triplicate. **P < 0.005 (parametric unpaired t test). (B) Split luciferase interaction assay was performed as in A, using Gluc1-FOS and Gluc2-JUN plasmids. The data shown are the mean ± SD of two independent experiments in triplicate. (C) Effect on the steady-state levels of the endogenous RED and SMU1 proteins. A549 cells were incubated with the indicated concentrations of LSP61 (+) or with DMSO (−) for 24 h. Total cell extracts were analyzed by Western blot, using antibodies specific for RED, SMU1, and GAPDH. (D) Cell viability assay. A549 cells were incubated with the indicated concentrations of LSP61 or with DMSO. ATP levels, which reflect the number of viable cells, were determined using the CellTiter-Glo Kit (Promega) at the onset of the experiment (T0) and following 24 h or 48 h of incubation (T24 and T48, respectively). The data shown are the mean of luciferase units ± SD of two independent experiments in duplicate. AU, arbitrary unit.

LSP61 Inhibits IAV Replication.

Finally, we assessed the ability of LSP61 to inhibit IAV replication. We first used the WSN-PB2-Nanoluc virus, in parallel with a recombinant human respiratory syncytial virus (RSV) expressing Firefly luciferase as a specificity control (RSV replicates in the cytoplasm with no direct involvement of the nuclear splicing machinery postulated). After viral adsorption to A549 cells, the LSP61 or LSP641 compound was added to the cells across the concentration range of 7.5–60 μM, and the luciferase activities were determined at 24 hpi. The RSV-Firefly signals showed only moderate reductions (Fig. 6A, gray bars). In contrast the WSN-Nanoluc signals showed a significant dose-dependent reduction, more pronounced for LSP61 (25-fold at 60 μM; P < 0.0001) compared with LSP641 (sixfold at 60 μM; P < 0.005) (Fig. 6A, black bars). Growth kinetics were performed with wild-type IAVs (the WSN virus and representatives of H1N1pdm09 and H3N2 circulating human IAVs) in the presence of 60 μM LSP61, and the supernatants were titrated by plaque assay. Because the H1N1pdm09 and H3N2 isolates grew poorly on A549 cells, canine MDCK-SIAT cells (32) were used in these experiments. In the presence of LSP61, the production of infectious particles showed a 10- to 100-fold reduction at 24 and 48 hpi (Fig. 6B). In comparison, in A549 cells, the production of WSN infectious particles was reduced about 1,000-, 100-, and 10-fold at 24, 48, and 72 hpi, respectively (Fig. 6C). In MDCK-SIAT cells, canine Mx proteins lack antiinfluenza activity, which limits the establishment of a strong IFN-induced antiviral state and favors a very efficient replication of influenza viruses (33). This particular feature of MDCK-SIAT cells likely accounts for the fact that a strong inhibition with the LSP61 compound was more difficult to achieve compared with that in A549 cells.

Fig. 6.

Evaluation of compound LSP61 for anti-IAV activity. (A) Effect on IAV replication. A549 cells were infected with the WSN-PB2-Nanoluc or RSV-Firefly virus [0.001 and 0.01 plaque forming units (PFU) per cell, respectively] and incubated with LSP641 or LSP61 at the indicated concentrations, or with DMSO. The luciferase units measured at 24 hpi are expressed as percentages relative to DMSO-treated control. The data shown are the mean ± SD of three (WSN-Nanoluc) or two (RSV-luc) independent experiments in triplicate. **P < 0.005, ***P < 0.0001 (parametric paired t test). (B and C) Effect on the production of infectious particles. MDCK-SIAT (B) or A549 (C) cells were infected with the WSN, H1N1pdm09, or H3N2 virus as indicated, and incubated with 60 μM LSP61 or with DMSO. At 24, 48, and 72 hpi, the supernatants were collected and viral titers were determined by plaque assay. The data are expressed as mean ± SD of three independent experiments [except for H1N1pdm09 and H3N2 at the 24 hpi time point in B (two independent experiments)], each in triplicate, that were pooled for titration. *P < 0.05, **P < 0.01, ***P < 0.001 (parametric paired t test). (D) Effect on NS1 mRNA splicing. A549 cells were infected with the WSN–wild-type (wt) virus (5 PFU per cell), and incubated with 60 μM LSP641 or LSP61, or with DMSO. After a 6-h incubation, poly(A)+ RNA were extracted, and the levels of NS2 and NS1 mRNAs were analyzed by quantitative real-time PCR and normalized with respect to GAPDH mRNA levels. The NS2/NS1 mRNA ratios shown are the mean ± SD of three independent experiments in duplicate. P = 0.01 (parametric paired t test).

In WSN-infected cells, the LSP641 and LSP61 compounds inhibited the splicing of the viral NS1 mRNA into NS2 mRNA at a concentration of 60 μM. Indeed, the NS2-to-NS1 mRNA ratio was reduced by 30% with LSP641, and by 50% with LSP61 (Fig. 6D), which is comparable to previous results obtained when either RED or SMU1 was knocked down in infected cells (23). In contrast, LSP641 and LSP61 did not inhibit the splicing of the viral M1 mRNAs (SI Appendix, Fig. S7A), in agreement with our previous findings that M1 mRNA splicing is little affected by RED or SMU1 silencing compared with NS1 mRNA splicing (23). Likewise, PB2 mRNA splicing, whose role in viral infection remains unclear (34), is not affected by 60 μM LSP641 or LSP61 (SI Appendix, Fig. S7B). The specificity of LSP641 and LSP61 toward the splicing of NS1 mRNA, taken together with our findings that both compounds inhibit RED–SMU1 interaction (Fig. 5A) and IAV replication (Fig. 6A), with the observed effects being consistently more pronounced for LSP61, strongly argue for their antiviral effect being due to disruption of the RED–SMU1 complex.

Discussion

We have determined the crystal structure of the interacting human SMU1_Nter and RED_mid. As expected from the high conservation of these two protein subdomains, the observed structure shows similarity to the one reported for C. elegans (29). In both structures, SMU1_Nter assembles into a dimer through intermolecular contacts between LisH motifs (SI Appendix, Fig. S2). Interestingly our structure reveals a more intricate SMU1–RED interaction with two binding interfaces, including a β-sheet/β-sheet interface (not present in the C. elegans structure) in addition to an α-helix/groove (SI Appendix, Fig. S3). Unlike Ulrich et al. (29), who assembled complexes from purified RED and SMU1 subdomains produced separately in Escherichia coli, we chose a coexpression strategy that allowed us to first delineate in HEK-293T cells and then to copurify from E. coli a very stable RED_mid–SMU1_Nter complex. RED and SMU1 are part of the spliceosomal precatalytic B complex, whose molecular architecture was very recently elucidated by cryo-EM (35). Fitting of our atomic structure of the RED_mid–SMU1_Nter complex enabled us to optimize the cryo-EM 3D model by applying the conformational constraint imposed by the newly revealed β-sheet/β-sheet interface between RED_mid and SMU1_Nter. The model (SI Appendix, Fig. S8) highlights a major structural role for RED, as it is positioned at the interface of the SMU1 LisH motif-based dimerization domain and the SF3B1 core spliceosomal factor. This pivotal position could contribute to a stabilization of the B complex structure during the early stages of spliceosome assembly that precede catalytic activation.

Based on our human RED_mid–SMU1_Nter complex structure, we designed pilot experiments and in silico screening, which led to the identification of two inhibitors, LSP641 and LSP61; they disrupt the RED–SMU1 interaction, reduce RED-SMU1 endogenous levels, inhibit IAV NS1 mRNA splicing and decrease the production of infectious IAV particles. These findings add to a growing list of more than 40 protein–protein interactions that have been successfully targeted with small drug-like molecules, at least in vitro (36, 37). Several inhibitors have entered clinical trials, including inhibitors of the p53–MDM2 interaction (38) and inhibitors of the interaction between the HIV-1 integrase and its cofactor LEDGF (39). Prospects for developing inhibitors seem to be higher for protein–protein interactions in which residues critical for the interaction are concentrated in small binding pockets, as is the case for the α-helix/groove interface between RED_mid and SMU1_Nter. In the recent years, much progress has also been made to enhance the therapeutic potential of short cell-permeable peptides as an alternative to conventional therapeutic small molecules (40). In principle, a peptide corresponding to the short RED[211–222] α-helix could be considered as a potential inhibitor of RED–SMU1 interaction. However, the strong hydrophobicity of this α-helix is likely to prevent its solubilization and to preclude such a strategy from being successful.

The potential toxicity associated with host-directed therapy also needs to be carefully evaluated. Cultured A549 cells treated with the RED-SMU1 targeting compounds LSP641 and LSP61 (this study), or treated with RED-SMU1 targeting siRNAs with a depletion efficiency of 90% (23), did not show cell death. Consistently, RED or SMU1 knockouts in C. elegans or Arabidopsis thaliana are viable (25, 27). However, a cytostatic effect of compound LSP61 was observed, as evidenced by bright-field imaging showing a lower density of the cell layer in the absence of dead cells and ATP quantification supporting a slower accumulation of metabolically active cells in culture wells. This observation is in agreement with previous reports showing that RED and SMU1 regulate alternative splicing of a subset of pre-mRNAs involved in development, apoptosis, and cell survival (25–27, 41, 42). The transcriptomic profiling of cells treated with compound LSP61 (or depleted for RED-SMU1 as a reference) will provide a means to investigate how LSP61 affects the expression and splicing of cellular genes, and to detect potential adverse effects to guide further drug development (43). Beyond their splicing function, RED and SMU1 are associated with the mitotic spindle (44) and chromatin (45), respectively, and are involved in the control of cell division (44, 46). The dual function of RED-SMU1 raises the question whether our observed antiviral effect of compounds LSP461 and LSP61 could be related not only to inhibition of viral mRNA splicing (as indicated by a reduced NS2-to-NS1 mRNA ratio upon treatment) but also to cell cycle arrest. Although this possibility cannot be formally excluded, it seems unlikely, as IAV infection per se has been shown to induce G0/G1 cell cycle arrest through inhibition of the RhoA/pRb signaling pathway (47) and down-regulation of cyclin D3 levels (48) in cultured cells. Besides, the tissue naturally targeted by IAVs (i.e., the airway epithelium) is essentially quiescent in healthy hosts (49). However, mitogenic stimulation through intratracheal administration of the keratinocyte growth factor in mice was found to induce the proliferation of alveolar type II cells (normally quiescent by more than 99%), and this increased their susceptibility to IAV infection (50). Nikolaidis et al. (50) suggest that the enhanced mortality due to influenza in infants or cigarette smokers could be related to a higher fraction of proliferating alveolar type II cells in these patients. Under this hypothesis, compounds targeting RED-SMU1 might show a double effect of inhibiting IAV replication through splicing modulation and limiting the intrapulmonary spread of IAV infection through cytostatic activity.

Over the past two decades, a growing number of small molecules that inhibit the spliceosome assembly or function have been identified, but many of them have yet to be characterized regarding their target and mechanism of action. The most documented are SF3B1 inhibitors of the spliceostatin, pladienolide, and sudemycin families (51–53). Although these drugs target a core splicing factor, there is evidence for a certain degree of selectivity. SF3B1 inhibitors affect splicing of pre-mRNAs associated with cell cycle regulation and apoptosis, and inhibit growth of cultured cancer cell lines at nanomolar concentrations, while cells derived from normal tissues are less sensitive (reviewed in refs. 54, 55). Preclinical studies have suggested that splicing modulation by SF3B1 inhibitors can be well tolerated in vivo and shows potential for the treatment of several types of cancers (56–58). Taking these data into account, together with the absence of observed cytotoxicity of LSP641 and LSP61 in our experiments, and with the fact that a topical treatment over a short time course can be considered in the case of acute IAV infections, compounds targeting the RED–SMU1 complex can reasonably be expected to show clinical safety and efficacy in influenza therapy. Our SMU1_Nter–LSP641 costructure provides a valuable basis for chemical optimization to select compounds with improved activity, specificity, and pharmacological properties, which can be tested in vivo in preclinical animal models for IAV infection. Such optimized RED-SMU1 disrupting compounds could be active against a wide range of IAVs, and less likely to select for resistance mutants. Interestingly, RED and SMU1 were identified as hits in two high-throughput screens looking for cellular factors involved in the HIV-1 life cycle (59, 60). SMU1 was also found to be associated with the E6 protein of the high-risk HPV18 human papillomavirus (61). Therefore, RED-SMU1 disrupting compounds might not only be active against IAVs but against a wider range of viruses that make use of the RED–SMU1 splicing complex. Besides potential applications as antivirals, our small molecule inhibitors might also serve as tools to dissect the early stages of spliceosome assembly, and lead to potential applications as anticancer agents (54, 62, 63).

Materials and Methods

Cells, Viruses, Plasmids, and Reagents.

Detailed information is provided in SI Appendix.

In Cellulo Split Luciferase-Based Interaction Assays.

The split-luciferase protein complementation assays were performed as described by Cassonnet et al. (64). Briefly, HEK-293T or A549 cells were seeded in 96-well white opaque plates (Greiner Bio-One) and cotransfected with 100 ng of each indicated pGluc1-P1 and pGluc2-P2 plasmid, where P1 and P2 represent proteins or protein subdomains of interest. The polyethylenimine PEI (Polysciences, Inc.) and JetPRIME (Polyplus Transfection) transfection reagents were used with HEK-293T and A549 cells, respectively. Normalized luciferase values were determined as described by Fournier et al. (23) and Cassonnet et al. (64). At 24 h posttransfection, the luciferase enzymatic activities were measured using the Renilla Luciferase Assay System (Promega) and a Berthold Centro XS luminometer. When cell-based interaction assays were performed for compound screening, A549 cells transfected with a full-length Gaussia luciferase expression plasmid were incubated in parallel with tested compounds or DMSO alone. The split-luciferase values were normalized with respect to control full-length luciferase signals.

Protein Expression in E. coli and Protein Purification.

The pETM11-SMU1_Nter plasmid was transformed in E. coli BL21-CodonPlus (Agilent). Cultures were grown at 37 °C in LB containing kanamycin (30 μg/mL) and chloramphenicol (34 μg/mL). When the OD₆₀₀ reached 0.6–0.8, the cultures were cooled down to 18 °C and expression was induced with 0.3 mM Isopropyl β-D-1-thiogalactopyranoside (IPTG). The cultures were incubated overnight at 18 °C before centrifugation. For the RED_mid–SMU1_Nter complex, a coexpression strategy was set up. The pETM11-RED_mid plasmid was transformed in E. coli BL21-RIL-SMU1_Nter–competent cells, and expression was performed using the protocol described above. Cell pellets were resuspended in lysis buffer [50 mM Hepes (pH 7.5), 150–500 mM NaCl, 1 mM β-mercaptoethanol (βme)] and sonicated. After centrifugation, the supernatant was complemented with imidazole to reach 25 mM and loaded on a nickel affinity resin (NiNTA, Qiagen). The resin was washed with a high-salt buffer [50 mM Hepes (pH 7.5), 1 M NaCl, 1 mM βme]. The recombinant proteins were eluted with elution buffer [50 mM Hepes (pH 7.5), 1 M NaCl, 1 mM βme, 300 mM imidazole]. Protein was dialyzed with tobacco etch virus protease overnight against the buffer without imidazole, loaded on a second Ni-NTA column, concentrated, and loaded on a size exclusion column: a HiLoad 16/600 Superdex 75 (GE Healthcare) for SMU1_Nter or a Superdex 200 Increase 10/300GL (GE Healthcare) for the RED_mid–SMU1_Nter complex. Fractions of interest were concentrated between 5 and 15 mg/mL.

Structure Determination.

For SMU1_Nter, the crystals (native and Se-Met derivative) were obtained in 0.1 M Bis-Tris (pH 5.5), 16–20% PEG 10K, and 0.1 M ammonium acetate, and cryoprotected in the same solution +30% glycerol. Data were processed with the XDS package (65). For experimental phasing of SMU1_Nter, a highly redundant single-wavelength anomalous dataset of an Se-Met–derived crystal was collected to 2.1-Å resolution at the peak of the Se-Met signal, as measured by X-ray fluorescence, on beamline ID29 (66) at the European Synchrotron Radiation Facility. For structural solution, eight Se-Met sites were located on the basis of their anomalous differences using SHELXC/D/E (67). These sites were subsequently refined, and experimental phases were calculated using the single anomalous dispersion procedure in SHARP (68). These phases were further improved by density modification, followed by model building with Buccaneer software (69). Model building and refinement were performed using the CCP4i suite program for crystallography (Phaser, ARP/wARP, REFMAC5, and Coot) (70–74).

For the RED_mid–SMU1_Nter complex, the crystals (native and Se-Met derivative) were obtained in 0.1 M Hepes (pH 7.0–7.5) and 8–10% PEG 8K, and cryoprotected in the same solution +30% glycerol. Data were processed with XDS, and the structure was solved by molecular replacement using the native SMU1_Nter structure. The model was built using the anomalous signal of an Se-Met derivative dataset for accurate side-chain attribution, and the final refinement was performed using the software cited above.

For the SMU1_Nter–LSP641 complex, a mother solution of LSP641 was prepared at 125 mM in DMSO. The cocrystals of SMU1_Nter (10 mg/mL) + 2.5 mM LSP641 were obtained in 0.1 M Bis-Tris (pH 6), 16–20% PEG 10K, and 0.2 M ammonium acetate, and cryoprotected in the same solution +30% glycerol. Data collection, processing, and model building were performed as described above.

In Silico Screening.

Docking experiments were performed using the LibDock protocol, as implemented in Discovery Studio (Discovery Studio Modeling Environment, release 4.5; Dassault Systèmes BIOVIA), an interface to the LibDock program developed by Diller and Merz (75). LibDock uses protein site features referred to as “HotSpots” that fall into two categories: polar and apolar HotSpots. The receptor HotSpot file was calculated before the docking procedure. Random ligand conformations were generated from the initial ligand structure through high-temperature molecular dynamics using a catalyst algorithm (76) before docking. The rigid ligand poses were placed into the active site, and HotSpots were matched as triplets. The poses were pruned, and a final optimization step was performed before the poses were scored. Ligand hydrogens, which were removed during the docking process, were added back to the ligand poses and optimized by minimization. The poses with the highest LibDock scores were retained and clustered according to their binding mode.

RNA Extraction and Reverse Transcription-Quantitative PCR.

Total RNA and poly(A)+ RNAs were extracted successively using the RNeasy Mini Kit (Qiagen) and Oligotex mRNA Mini Kit (Qiagen) following the manufacturer’s protocol. Reverse transcription (RT) was performed on 5–10 ng of poly(A)+ RNA using a Maxima First Strand cDNA Synthesis Kit (Thermo Fisher Scientific). Quantitative real-time PCR was performed on 2 μL of a 1/10 dilution of the RT reaction using Luminaris Color Probe qPCR Master Mix (Thermo Fisher Scientific) and a Light Cycler 480 (Roche). The levels of NS1/NS2 and M1/M2 mRNAs were determined and normalized with respect to GAPDH mRNA levels using the protocol described by Fournier et al. (23). The levels of PB2/PB2-S1 mRNAs were determined by semiquantitative RT-PCR as described by Yamayoshi et al. (34).

Data Availability Statement.

Structure coordinates and diffraction data are available at the Protein Data Bank (PDB) (https://www.rcsb.org/) under PDB ID codes 6Q8F (SMU1_Nter), 6Q8I (RED_mid-SMU1_Nter), and 6Q8J (SMU1_Nter-LSP641).