Xinqidi,Your biological research partner
Xinqidi Biotech Co.,Ltd,Wuhan,China 2008-2021
R&D 13th year

Structure of PDE3A-SLFN12 complex reveals requirements for activation of SLFN12 RNase

Issuing time:2021-07-22 11:13


DNMDP and related compounds, or velcrins, induce complex formation between the phosphodiesterase PDE3A and the SLFN12 protein, leading to a cytotoxic response in cancer cells that express elevated levels of both proteins. The mechanisms by which velcrins induce complex formation, and how the PDE3A-SLFN12 complex causes cancer cell death, are not fully understood. Here, we show that PDE3A and SLFN12 form a heterotetramer stabilized by binding of DNMDP. Interactions between the C-terminal alpha helix of SLFN12 and residues near the active site of PDE3A are required for complex formation, and are further stabilized by interactions between SLFN12 and DNMDP. Moreover, we demonstrate that SLFN12 is an RNase, that PDE3A binding increases SLFN12 RNase activity, and that SLFN12 RNase activity is required for DNMDP response. This new mechanistic understanding will facilitate development of velcrin compounds into new cancer therapies.


A class of small molecules has recently been described that causes selective cancer cell killing by inducing complex formation between two cellular proteins, PDE3A and SLFN121,2,3. These small molecules are exemplified by the prototypical PDE3A-SLFN12 complex inducer, DNMDP1, although other classes of PDE3A-SLFN12 complex inducers have subsequently been described with similar, albeit weaker, activity4,5,6,7. Cancer cells expressing elevated levels of both PDE3A and SLFN12 are typically sensitive to killing by DNMDP and other PDE3A-SLFN12 complex inducers.

PDE3A is a well-characterized cyclic nucleotide phosphodiesterase (PDE) that hydrolyzes cAMP, cGMP, and cUMP8,9. DNMDP is a PDE3A inhibitor and the catalytic domain of PDE3A is sufficient to support response to DNMDP1,3. However, the cancer cell killing activity of DNMDP does not correlate with inhibition of PDE3A enzymatic activity, in that other potent and selective PDE3 inhibitors such as trequinsin10 do not kill cancer cells, and knockout of PDE3A from sensitive cell lines abolishes DNMDP sensitivity1,3. DNMDP instead has a gain- or change-of-function effect on PDE3A, involving induction of complex formation with SLFN12.

Unlike PDE3A, little is known about the normal physiological function of SLFN12 beyond its expression in T cells and association of ectopic expression with differentiation and/or quiescence11,12,13,14. Most human Schlafen genes, including SLFN5, SLFN11, SLFN13, and SLFN14, are significantly longer than SLFN12 and encode a C-terminal helicase domain and a nuclear localization signal, which SLFN12 and the related SLFN12L do not encode15. The six human SLFN genes have a semi-conserved SWADL motif in common16, as well as a divergent AAA ATPase domain that may function as an RNA-binding domain17.

We took multiple orthogonal approaches to determine how DNMDP-like molecules induce PDE3A-SLFN12 complex formation. We first solved the crystal structure of the PDE3A catalytic domain bound to several ligands, including DNMDP and trequinsin. This structure was used to map intramolecular and intermolecular changes in PDE3A upon complex formation with SLFN12 using hydrogen–deuterium exchange mass spectrometry (HDX-MS). The full cryo-electron microscopy (Cryo-EM) structure of the PDE3A-SLFN12 complex with bound DNMDP revealed the molecular details of complex formation and the role of DNMDP in stabilizing the complex. Deep mutational scanning (DMS) of PDE3A identified amino acids required for DNMDP response, including residues involved in compound binding, PDE3A homodimerization, and SLFN12 binding, substantiating findings from the structural studies. Finally, we show that, like other SLFN family members18,19,20, SLFN12 is an RNase, that RNase activity is increased upon PDE3A binding, and that SLFN12 RNase activity is required for DNMDP response.


DNMDP induces complex formation between purified PDE3A and SLFN12

Previous work showed that DNMDP induces complex formation between PDE3A and SLFN12 in cells1. To determine whether this complex can be recapitulated with isolated proteins, we expressed and purified the catalytic domain of PDE3A (PDE3ACAT), which comprises residues 640–1141, and full-length SLFN12 and analyzed their ability to interact in the absence and presence of DNMDP. We limited our analysis to the catalytic domain of PDE3A because our previous experiments indicated that the N-terminal portion of PDE3A, containing several membrane association domains21, was not required for DNMDP sensitivity in cells3. Analysis of recombinant PDE3ACAT by analytical size exclusion chromatography (SEC) revealed a single species at an elution volume earlier than expected for a monomer of theoretical mass 57.3 kDa (Fig. 1a). The solution mass of PDE3ACAT was determined to be 118 kDa using multi-angle light scattering (MALS) as the protein eluted from a size exclusion column (SEC-MALS), showing that the catalytic domain existed as a dimer (Supplementary Fig. 1). Dimerization of PDE3ACAT was supported by sedimentation equilibrium analytical ultracentrifugation (SE-AUC) data, which could be fit to a monomer–dimer equilibrium with a Kd of 40 nM (Fig. 1b). Considering that PDE3A contains additional regions N-terminal to the catalytic domain that are proposed to promote oligomerization21, it is likely that the Kd for dimerization of full-length PDE3A protein is even lower than observed here. Purified SLFN12 showed concentration-dependent aggregation at low NaCl concentrations and was stored and analyzed at 500 mM NaCl, except where noted. Analysis by SEC revealed that SLFN12 eluted significantly later than would be expected for a protein of theoretical mass 67.3 kDa, suggesting a non-specific interaction with the column resin (Fig. 1a). Measurement of the solution mass of SLFN12 by SEC-MALS gave a mass of 126.9 kDa, showing that it also existed as a dimer at micromolar protein concentrations (Supplementary Fig. 1). SE-AUC analysis of SLFN12, however, showed that the dimer was significantly less stable than observed for PDE3ACAT, with the data fitting to a monomer–dimer Kd of 870 nM (Fig. 1c).

Fig. 1: DNMDP induces complex formation between purified PDE3A and SLFN12.

a SEC analysis of complex formation performed at 150 mM NaCl. Traces are shown for 10 µM PDE3ACAT (red), SLFN12 (green), and PDE3ACAT + SLFN12 in the absence (blue) and presence of 100 µM DNMDP (black). The trace for SLFN12 had to be collected at 500 mM NaCl and is therefore shown as a dashed line. b SE-AUC analysis of PDE3ACAT and c SLFN12. The solid lines represent the best fit to a monomer–dimer model with the Kd indicated. d Amylose resin pulldown analysis of complex formation. SDS-PAGE gel of protein(s) eluted from amylose resin when 2 µM of His6-MBP-SLFN12 was incubated alone or with 2 µM PDE3ACAT ± 10 µM DNMDP or trequinsin at 150 or 500 mM NaCl. e SEC analysis of complex formation performed at 500 mM NaCl. f Effect of DNMDP and trequinsin on complex formation by BLI. Binding of 100 nM SLFN12 to immobilized PDE3ACAT in the presence of 500 mM NaCl and DMSO (blue), 10 µM DNMDP (black), or 10 µM trequinsin (red). gQuantitative analysis of the effect of DNMDP on complex formation using BLI. Binding to SLFN12 performed in the absence (open circles) and presence (filled circles) of 100 µM DNMDP at 500 mM NaCl. Data were analyzed from four independent experiments and represented as mean values ± s.e.m. Abs. absorbance, Treq. trequinsin.

To investigate the interaction between PDE3A and SLFN12, a pulldown assay was performed with a version of SLFN12 with maltose-binding protein (MBP) fused to its N-terminus. MBP-SLFN12 was immobilized on amylose resin and incubated with PDE3ACAT in the absence and presence of DNMDP at 150 mM NaCl (Fig. 1d). Under these conditions, PDE3ACAT was able to bind MBP-SLFN12 in the absence of compound, and the quantity bound was only slightly increased in the presence of DNMDP. In contrast, trequinsin, a potent and selective PDE3A/B inhibitor (Supplementary Table 1) with no cancer cell killing activity1, inhibited binding of PDE3A to MBP-SLFN12. Curiously, when the experiment was repeated at 500 mM NaCl, less PDE3ACAT interacted with MBP-SLFN12 except in the presence of DNMDP, which dramatically increased complex formation (Fig. 1d).

SEC was used to further investigate complex formation between PDE3A and SLFN12 and the effect of DNMDP. At physiological salt concentration (150 mM NaCl), the two proteins coeluted at an earlier volume than either protein alone, indicative of stable complex formation (Fig. 1a). Addition of DNMDP had only a minor effect on complex formation under these conditions. However, at 500 mM NaCl, the earlier eluting complex was stably formed only in the presence of DNMDP (Fig. 1e). In contrast, the chromatogram of the PDE3A-SLFN12 run in the absence of DNMDP was consistent with partial dissociation into constituent PDE3A and SLFN12 proteins. When measured by SEC-MALS, the mass of the PDE3ACAT + SLFN12 complex at 150 mM NaCl was determined to be 246.9 kDa, in close agreement with a dimer of PDE3ACAT interacting with a dimer of SLFN12 (Supplementary Fig. 1).

To quantify the effect of DNMDP on complex formation, we immobilized biotinylated Avi-tagged PDE3ACAT on a streptavidin biosensor and measured its interaction with SLFN12 using biolayer interferometry (BLI; Fig. 1f, g). These experiments could only be performed at 500 mM NaCl as SLFN12 showed non-specific binding to the sensors at 150 mM. We observed clear association between SLFN12 and the immobilized PDE3A. Addition of DNMDP increased the rate of association and reduced the rate of dissociation of SLFN12 with PDE3A. The very slow dissociation rate suggests that the complex is highly stable over time. Trequinsin, as expected, inhibited binding. The calculated steady state Kd for complex formation at 500 mM NaCl in the absence and presence of DNMDP was 320 and 65 nM, respectively (Fig. 1g).

Taken together, these data demonstrate that purified PDE3ACAT and SLFN12 are sufficient to form a stable complex comprised of a dimer of PDE3A and a dimer of SLFN12 and that DNMDP significantly stabilizes the complex. The co-chaperone, aryl hydrocarbon receptor interacting protein (AIP), is required for PDE3A-SLFN12 complex formation in cells3. However, AIP is evidently not required for complex formation in vitro. The influence of salt concentration on complex formation suggested that electrostatic interactions play an important role in promoting complex formation.

DNMDP and trequinsin have only a limited effect on the structure of PDE3ACAT

DNMDP was previously shown to be an inhibitor of PDE3A PDE activity, implying that DNMDP binds directly to PDE3A1. Indeed, the melting temperature (Tm) of PDE3ACATincreased by +3 °C in the presence of DNMDP (Supplementary Fig. 1), indicating that DNMDP binds and stabilizes the structure of PDE3ACAT. In contrast, no change in Tm was observed upon incubating SLFN12 with DNMDP (Supplementary Fig. 1). This suggests that SLFN12 does not bind to DNMDP in the absence of PDE3A, although we cannot discount the possibility that its binding does not sufficiently impact the stability of the structure to cause a change in the melting temperature of SLFN12.

To gain insight into how DNMDP induces complex formation of PDE3A with SLFN12 and how trequinsin inhibits this interaction, we solved the high-resolution crystal structure of a modified form of the catalytic domain of PDE3A (PDE3ACAT-Xtl) in the absence or presence of DNMDP or trequinsin (Fig. 2a and Supplementary Table 2). PDE3ACAT-Xtl is comprised of residues 669–1095 with two internal loops between residues 780–800 and 1029–1067 replaced with shorter linkers to aid in crystallization and improve diffraction quality of the crystals (Fig. 2a). We also obtained the structure of AMP bound to PDE3A by soaking the apo-crystals with cAMP, demonstrating that the crystalline form of PDE3ACAT-Xtl was catalytically active, similar to PDE4D2 crystallized with cAMP22. The catalytic domain of PDE3A crystallized as a dimer with two dimers in the asymmetric unit (Fig. 2a). The structure is very similar to the structure of the catalytic domain of PDE3B23, with which it shares 69% sequence identity, with a root mean square deviation (RMSD) of 0.53 Å for the main chain atoms. The PDE3A crystal structures are essentially identical, with only small movements of the protein backbone at the active site to accommodate the binding of AMP, DNMDP, and trequinsin (Supplementary Fig. 1).

Fig. 2: DNMDP has only a limited effect on the structure of PDE3ACAT but HDX-MS reveals regions of PDE3ACAT with decreased solvent exposure following SLFN12 binding.

a Crystal structure of PDE3ACAT-Xtl bound to DNMDP. Each monomer is colored dark and light blue. DNMDP is shown in space-filling for clarity with the carbon, oxygen, and nitrogen atoms colored green, red, and blue, respectively. The two loop regions replaced by linkers are indicated by dashed lines. bd Catalytic sites from the PDE3ACAT-Xtl crystal structures. b PDE3ACAT-Xtl-AMP; c PDE3ACAT-Xtl-DNMDP; and d PDE3ACAT-Xtl-trequinsin are shown. The side chains are shown in a licorice format, with the carbon atoms colored white. The AMP, DNMDP, and trequinsin are shown in a similar format except the carbons that are colored green for visualization purposes. The phosphorus of AMP is shown in orange. The two metal ions and the coordinating water molecules are represented as gray and red spheres, respectively. Hydrogen bonds and metal–ligand interactions are shown as dashed yellow lines. e Analysis of the PDE3ACAT and SLFN12 interface by HDX-MS. D-uptake differences are mapped with green and blue onto the PDE3ACAT-Xtl crystal structure for 300 and 3000 s. BRD9500 is depicted with cyan spheres.

In the AMP-bound structure, the AMP is stretched across the catalytic site, with the phosphate group coordinating directly with the two metal ions at one end and the purine moiety making hydrogen bonds with Q1001 at the other end (Fig. 2b). The adenine moiety is further stabilized by hydrophobic contacts with I968 and ππ stacking with F1004. In the DNMDP-bound structure, DNMDP binds in an extended conformation with the dihydropyrazidinone ring buried deep in the active site pocket (Fig. 2c). This orientation is stabilized by hydrogen bonds with H961 and Q1001 and a series of hydrophobic interactions along the length of DNMDP. F972 and L910 also pack against each of the ethyls of the diethylamino group of DNMDP, which are positioned at the entrance of the active site pocket. Trequinsin binds in a similar location as DNMDP, with the aromatic ring at the di-methoxy end being roughly superimposable on the phenyl ring of DNMDP; however, it is not as deeply embedded into the active site and is stabilized only by non-polar interactions (Fig. 2d). The trimethylphenyl group of trequinsin, which is rotated 90° relative to the plane of the rest of the molecule, additionally packs into a hydrophobic pocket created by the Cβ of S1003 and several hydrophobic side chains at the entrance of the catalytic site.

It is clear from the crystal structures that DNMDP and trequinsin inhibit PDE3A catalytic activity by sterically restricting entry of cAMP and preventing key stabilizing contacts. However, they do not reveal how DNMDP promotes SLFN12 binding, as DNMDP does not cause any obvious structural changes in PDE3A. We hypothesized that the interaction with SLFN12 is stabilized by contacts with the exposed diethylamino group of DNMDP and inhibited by the trimethylphenyl group of trequinsin.

HDX-MS identified three regions of PDE3ACAT with decreased solvent exposure following SLFN12 binding

To gain further insight into PDE3A-SLFN12 complex formation, we utilized HDX-MS to identify regions of PDE3A that were affected upon binding to SLFN12. HDX-MS provides an excellent way to probe protein–protein interactions in solution by comparing the relative deuterium uptake between amide hydrogens of the protein backbone in the protein alone and bound to a partner24. Experiments were performed with PDE3ACAT bound to the DNMDP analog BRD95002 in the absence and presence of SLFN12, using a high salt concentration (500 mM) to ensure a structurally uniform SLFN12 population.

Analysis of the uptake of deuterium into PDE3ACAT in the presence of BRD9500 revealed many slowly exchanging regions, which corresponded to well-folded regions of the protein observed in the crystal structure (Supplementary Fig. 2 and Supplementary Data 1). Extensive deuteration at the earliest time point was observed in both the N- and C-terminals of the protein and in the loop regions between 778–793 and 1028–1068, which indicates dynamic regions.

The deuterium uptake profiles for PDE3ACAT bound to BRD9500 were then compared to data collected in the presence of SLFN12, focusing on the regions that show either an increase or decrease in deuterium exchange. In the presence of SLFN12, three distinct regions of PDE3ACAT exhibited a decrease in exchange of deuterium: region 1 covering residues 849–867, region 2 covering residues 902–940, and region 3 covering residues 983–1001 (Fig. 2e, Supplementary Fig. 2, and Supplementary Data 1). A decrease in deuterium exchange is often associated with the region becoming shielded from the solvent, potentially as a result of interacting with a partner protein. This is likely the case with solvent-exposed regions 2 and 3. Intriguingly, they are also at or near the DNMDP-binding site, supporting the hypothesis that SLFN12 binds close to this region. Region 1, which encompasses peptides at the PDE3ACAT dimerization interface, exhibited the greatest decrease in deuterium exchange. This suggests that the binding of SLFN12 stabilizes the PDE3ACAT homodimer further, presumably via the reduction of structural fluctuations in the dimer interface, not captured in the crystal structure. Under the experimental conditions employed, we did not detect any long-distance allosteric changes in PDE3A upon interaction with SLFN12, presumably with regions 2 and 3, although the deuterium changes detected in the PDE3A dimer interface may be considered such.

Cryo-EM solution of the PDE3A-SLFN12 complex structure

To directly address how DNMDP promotes PDE3A-SLFN12 complex formation, we used Cryo-EM to solve the structure of the PDE3ACAT-DNMDP complex bound to full-length SLFN12 (Supplementary Fig. 3 and Supplementary Table 3). The complex was found to possess twofold symmetry, which was applied throughout refinement, and consisted of two flexibly connected bodies (Supplementary Movie 1). The bodies could be readily identified by fitting with the PDE3ACAT-Xtl dimer crystal structure and two monomers of the N-terminal domain (NTD) of rat SLFN13, which shares 38% sequence identity with the related region of SLFN1220. The dynamic motion of the two bodies necessitated refinement of each protein dimer separately utilizing the multi-body refinement approach in RELION25, with the final resolution for the maps for SLFN12 and PDE3ACAT being 2.76 and 2.97 Å, respectively (Supplementary Figs. 3 and 4). Residues 669–1100 for each monomer of PDE3ACAT could be modeled except for the loop regions between 779–799 and 1029–1068, which likely adopted multiple conformations. These loop regions are equivalent to the ones that were replaced by short linkers in the crystal structure. Residues 1–560 of each monomer of SLFN12 could be modeled, with the first 346 residues being guided by the structure of rat SLFN13 and residues 387–560 built de novo. The regions between 346 and 386 and beyond residue 561 were not visible in the maps, suggesting that they adopted multiple conformations. There was also clear density for two DNMDP molecules, one in the catalytic site of each PDE3A monomer (Supplementary Fig. 4).

In agreement with the biophysical studies, the Cryo-EM structure shows a dimer of PDE3ACAT-DNMDP interacting with a dimer of SLFN12 (Fig. 3a). The structure of PDE3ACAT is essentially the same as PDE3CAT-Xtl (RMSD for backbone atoms of 0.38 Å), with only an additional two turns of the C-terminal helices for each monomer modeled (Supplementary Fig. 5). There was no density evident for the loop regions between residues 779–799 and 1029–1068, suggesting that they adopt multiple conformations and are not involved in contacting SLFN12. In the HDX-MS studies, these regions showed a high deuterium content at the earliest time point in the absence of SLFN12, indicating that they are dynamic (Supplementary Fig. 2). The deuterium uptake did not change in the presence of SLFN12, supporting the observation that they are not affected by complex formation (Supplementary Fig. 2).

Fig. 3: Overview of the Cryo-EM structure of the PDE3ACAT and SLFN12 complex.

a Structure of the PDE3ACAT-SLFN12-DNMDP heterotetramer. PDE3ACAT (dark and light blue) and SLFN12 (green and magenta) are shown in a surface and cartoon representation, respectively. The loop regions between 779–799 and 1029–1068, which do not show density in the Cryo-EM maps, are indicated by dashed lines on one PDE3A monomer (light blue). The zinc ions of SLFN12 are shown as gray spheres. b Detailed view of the SLFN12 monomer structure. A schematic showing the different regions of SLFN12 is shown at the top with the structure of the dimer shown below. One monomer is shown in a surface representation (white) and the other in cartoon representation with different regions color-coded to match the schematic. Residues that are involved in stabilizing contacts within the C-terminal domain and between the N- and C-terminal domains (F548, A12, and P519) are indicated. BD bridging domain. c Summary of the SLFN12 dimer interface interactions. Each of the points of contact is denoted by a key residue, which is shown in the top left corner of each panel. The backbone of each monomer, and the labels of the residues shown, are colored green or magenta depending on which monomer they come from. The labels for the monomer in magenta also have a prime mark to indicate that they are from the other monomer.

The structure of the SLFN12 monomers can be divided into an NTD and C-terminal domain (CTD), which comprise residues 1–345 and 387–560, respectively (Fig. 3b). The SLFN12-NTD can be further sub-divided into an N-lobe, C-lobe, and two bridging domains (BDs; Fig. 3band Supplementary Fig. 5). The overall structure of the SLFN12-NTD is very similar to the rat SLFN13-NTD, with an RMSD of main chain atoms of 0.55 and 0.75 Å for the individual N-lobes and C-lobes, respectively. The SLFN12-CTD is comprised of a core domain between residues 387–541 and the PDE3A interacting region (PIR), between residues 551 and 560 (Fig. 3b and Supplementary Fig. 5). The linker between these two regions gives rise to the flexibility observed between the PDE3A and SLFN12 dimers. This linker is stabilized by hydrophobic interactions between F548, other residues in the linker, and the CTD core region (Fig. 3b and Supplementary Fig. 5). While the SLFN12-NTD and -CTD are connected by a flexible linker, they are stabilized relative to each other by a series of hydrophobic and salt bridge interactions (Supplementary Fig. 5). The two SLFN12 monomers interact over the length of the NTD and a small portion of the CTD (Fig. 3b). The interactions are centered around four residues from each monomer: T71, F89, I131, and I517. The side chains of these four amino acids are inserted into hydrophobic pockets formed by residues from the opposite monomer (Fig. 3c).

The C-terminal helix of SLFN12 is primarily responsible for complex formation

The majority of the interactions between PDE3ACAT and SLFN12 occurs via the PIR sequence 551-AENLYQIIGI-560 at the C-terminal end of each SLFN12 monomer, which extends out from the main body of SLFN12 (Figs. 3 and 4a). Almost all of the amino acids in this sequence interact directly with residues around the entrance of the catalytic site of PDE3A through hydrophobic interactions (Fig. 4b). This includes a close packing interaction between one of the ethyls of the diethylamino group of DNMDP and I557 of SLFN12, which provides a direct explanation for the increase in stability of the PDE3A-SLFN12 complex in the presence of DNMDP (Fig. 4b). It is interesting to note that several PDE3A residues are involved in contacting both DNMDP and SLFN12 (Supplementary Table 4). It is possible that the binding of DNMDP may help stabilize the position of these residues to optimize interaction with SLFN12. Modeling the position of trequinsin into the Cryo-EM structure, based on its location in the crystal structure, clearly shows that it would sterically clash with the L554, I557, and I558 side chains of SLFN12, making the interaction between PDE3A and SLFN12 highly unstable (Fig. 4c). Outside of the C-terminal region of SLFN12, the only other point of contact between the two proteins occurs through a short loop from each PDE3A monomer that contains two acidic residues, D926 and D927, which point toward a highly positively charged surface of SLFN12 (Fig. 4a, d). D927 from PDE3A forms a direct salt bridge with K150 from SLFN12 from each monomer at this interface (Fig. 4a). This may in part explain why binding between PDE3A and SLFN12 was destabilized at higher salt concentrations.

Fig. 4: Intermolecular interactions between PDE3ACAT and SLFN12.

a Summary view of intermolecular interactions between PDE3ACAT and SLFN12. The surface representation around D927 was made transparent to highlight the side chains. The residues are labeled in the color of the protein monomer they come from. b Interactions between the C-terminal region of SLFN12 and PDE3A. The SLFN12 residues that are the focus of each panel are shown in the upper left corner. The backbone and labels for the side chains of PDE3A are shown in blue and SLFN12 in green. c Modeling the effect of trequinsin on PDE3ACAT-SLFN12 complex formation. Trequinsin, shown in a thin stick format and colored per atom to differentiate from DNMDP (blue), is modeled based on superimposing the PDE3ACAT-Xtl-trequinsin structure onto PDE3ACAT in the Cryo-EM structure. d Distribution of surface charges on SLFN12. For the last figure, the location of the two putative RNAse catalytic residues on each monomer are indicated. The surface charge for the SLFN12 dimer was calculated using APBS47 and visualized in PyMOL. The surface was colored with a gradient going from blue (highly positively charged) to white (neutral), to red (highly negatively charged).

The importance of the SLFN12 PIR to PDE3A-SLFN12 complex formation was supported by deletion studies. Whereas a 10-amino acid C-terminal truncation did not impair DNMDP response when transduced into HeLa-Res cells lacking endogenous SLFN12 expression, a 30-amino acid truncation removing the PIR domain, but with similar levels of expression, completely abolished DNMDP response (Fig. 5a–c). These results demonstrate a requirement of the SLFN12 PIR for response to DNMDP.

Fig. 5: The PIR helix of the SLFN12 CTD is required for DNMDP-induced cancer cell killing.

a Schematic diagram of SLFN12 deletion mutants. b Seventy-two-hour DNMDP-response viability assay of SLFN12 truncation mutants deleting the PIR region expressed in HeLa-Res cells lacking endogenous SLFN12 expression. Data are plotted as mean values with error bars indicating +/−standard deviation of four replicates. c Anti-flag immunoblot analysis showing the expression levels of flag-tagged SLFN12 proteins, marked with arrows.

Deep Mutational Scanning of PDE3A identifies DNMDP resistance mutations

To further investigate the structural relationship between PDE3A, DNMDP, and SLFN12, we used DMS to identify residues of PDE3A that impact DNMDP sensitivity. Because we previously showed that the isolated catalytic domain of PDE3A was sufficient to confer DNMDP sensitivity in cells expressing SLFN12 but lacking endogenous PDE3A3, we limited our mutational analysis to the PDE3A catalytic domain. We designed a library of PDE3A alleles in which the sequence encoding amino acids 668–1141, including the catalytic domain, was substituted with a codon for every other possible amino acid or a stop codon in the context of the full-length cDNA (Supplementary Data 2). This library was expressed in PDE3A-knockout GB1 glioblastoma cells (Supplementary Fig. 6) and assessed for survival in the presence of dimethyl sulfoxide (DMSO), 100 nM DNMDP, or 100 nM trequinsin. After elimination of highly variable sites due to low representation in the PDE3A DMS library (Supplementary Fig. 6), we compared survival results from cells treated with 100 nM DNMDP or DMSO (Fig. 6a).

Fig. 6: Deep mutational scanning of PDE3A identifies DNMDP resistance mutations.

a Average log2 fold changes (LFC) in abundance of mutant alleles of PDE3A following treatment with DNMDP as compared to treatment with DMSO. Wild-type (wt) alleles are marked with silent single-nucleotide polymorphisms (SNPs) (purple), whereas stop codons (red) result in complete loss of function until the last 59 amino acids. One standard deviation of increased survival is indicated with the dotted line. Resistance mutations with a survival score greater than one standard deviation from the mean and that are not predicted to disrupt PDE3A protein folding are colored to indicate localization to the active site (orange), homodimerization domain (green), or putative binding site of SLFN12 (dark blue). b Location of resistance mutations mapping to the PDE3A active site (orange), PDE3A homodimer interface (green), and putative SLFN12-binding site (dark blue). The backbone of the two PDE3A monomers are shown in light blue and white, with the mutations mapped onto the white monomer. The DNMDP is represented as space-filling and colored cyan. cPulldown of wild-type or mutant PDE3A with a resin-conjugated DNMDP analog following ectopic expression in PDE3A-knockout A2058 cells. PDE3A proteins (marked with arrows) recovered in the pellet (P lanes) were visualized by western blotting using a PDE3A antibody. The presence of 10 μM trequinsin (T lanes) during pulldown competed away the binding of PDE3A to the resin, resulting in a negative pulldown. d Co-immunoprecipitation of ectopically expressed wild-type or mutant PDE3A with flag-tagged wild-type SLFN12 following ectopic expression in PDE3A-knockout HeLa cells. SLFN12 proteins, marked with arrows, that co-immunoprecipitated with PDE3A were detected by anti-FLAG western blotting. e Confirmation of deep mutation scanning results with selected PDE3A mutants in PDE3A-knockout A2058 cells. Seventy-two-hour Cell-Titer Glo assay. GFP, green fluorescent protein. f Dimer interface in the PDE3ACAT-Xtl structure. The labels for the residues are colored light and dark blue depending on which monomer they come from.

The majority of mutations that enabled GB1 cell survival mapped to residues contributing to the stability of the protein fold. Remaining loss-of-function mutations that enabled GB1 cell survival in the presence of DNMDP fell largely into two categories: (i) mutations that cluster around or in the catalytic site and (ii) mutations that surround the PDE3A homodimerization domain (Fig. 6b). These mutations permit GB1 survival in the presence of DNMDP, with a log2 fold change (LFC) >0, similar to that supported by introduction of stop codons. We previously reported that mutations of the PDE3A active site abolished DNMDP binding and cytotoxic response3. The DMS and HDX results also implicate the dimerization interface of PDE3A as having a key role in PDE3A-SLFN12 complex formation. Supporting these results, mutation of the PDE3A homodimerization domain, exemplified by N867R, did not prevent compound binding (Fig. 6c) but did inhibit PDE3A-SLFN12 complex formation and DNMDP response (Fig. 6d–f). Interestingly, several substitutions of F914, in a region of PDE3A implicated in SLFN12 binding by HDX and Cryo-EM, also caused resistance to DNMDP. We therefore individually assessed two mutations of F914 and found that F914A and F914D retained the ability to bind resin-conjugated compound1, albeit with decreased efficiency (Fig. 6c), but no longer complexed with SLFN12 in response to DNMDP (Fig. 6d) and failed to support DNMDP cytotoxic response (Fig. 6e). Mutation of several other PDE3A residues determined by Cryo-EM to be located at the PDE3A–SLFN12 interface also caused resistance to DNMDP (Supplementary Table 4 and Supplementary Data 2).

SLFN12 has RNase activity, which is required for DNMDP-induced cell killing

SLFN13 was recently reported to contain an N-terminal RNase domain20, a region with 35% sequence identity to SLFN12 (Supplementary Fig. 7). We hypothesized that SLFN12 is also an RNase and that this activity may contribute to DNMDP-induced cell killing. To test our hypothesis, we mutated two residues of SLFN12, E200 and E205, orthologous to the active site residues of SLFN13 (Supplementary Figs. 5 and 7), and analyzed the effect of these mutations on DNMDP-induced complex formation and cell killing. The SLFN12 mutants all interacted with endogenous PDE3A in HeLa cells upon treatment with DNMDP (Fig. 7a), suggesting that the mutations did not interfere with complex formation. However, whereas ectopic expression of wild-type SLFN12 conferred DNMDP sensitivity in HeLa-Res cells lacking endogenous SLFN12 expression, ectopic expression of mutant SLFN12 did not, suggesting a requirement for an E200/E205-dependent SLFN12-intrinsic enzymatic activity for response to DNMDP (Fig. 7b).

Fig. 7: SLFN12 RNase activity is required for sensitivity to DNMDP.

a Complex formation of PDE3A with predicted SLFN12 catalytic mutants was assessed by ectopically expressing V5-tagged wild-type or mutant SLFN12 constructs in HeLa cells, treating with DNMDP for 8 h, immunoprecipitating endogenous PDE3A, and detecting coprecipitated SLFN12 by V5 immunoblot. b DNMDP sensitivity of SLFN12 cDNAs with mutations of predicted RNase catalytic residues in HeLa-res cells lacking endogenous SLFN12 expression. Data are plotted as mean values with error bars indicating +/−standard deviation of four replicates. c Direct measurement of SLFN12 RNase activity. In all, 2 µM of wild-type (WT) or the active site mutant (E200A, E205A) SLFN12 were incubated with 2 µg of human rRNA at 37 °C for 40 min. Wild-type SLFN12 was denatured (WT, Den) by heating at 65 °C for 20 min. Cleaved rRNAs were analyzed on a formaldehyde agarose gel. dInduction of SLFN12 RNase activity by PDE3A-SLFN12 complex formation. 0.25 µM PDE3A and SLFN12 proteins were incubated with DMSO, 12.5 µM DNMDP, 12.5 µM trequinsin, or 12.5 µM estradiol at room temperature for 30 min prior to a 1:10 dilution of complex and incubation with rRNA. Digested rRNAs were analyzed on a denaturing agarose gel. The relative amount of intact 28S rRNA was quantified using the ImageJ software, shown at the bottom of the figure. To detect the complex formation between PDE3A and SLFN12, preincubated recombinant proteins were crosslinked using 1-ethyl-3-(3-dimethylaminopropyl)carbodiimide (EDC) and N-hydroxysuccinimide (NHS). Reaction was stopped by adding SDS loading buffer. Proteins were separated on an SDS-PAGE gel and detected by silver staining. e RNase activity of the isolated SLFN12 NTD. Experiment with SLFN12 amino acids 1–347 performed as in d.

To determine whether SLFN12 is indeed an RNase, we incubated 2 µM recombinant SLFN12 with human rRNA isolated from HeLa cells in a buffer containing 40 mM Tris-HCl, pH 8.0, 20 mM KCl, 4 mM MgCl2, and 2 mM dithiothreitol (DTT). At this protein concentration, most of the SLFN12 protein would be expected to exist in a dimeric form. Under these conditions, wild-type SLFN12, but not heat-denatured SLFN12, was able to degrade the rRNA (Fig. 7c). This RNase activity was largely inhibited by mutation of E200 or E205. We repeated this experiment with 0.25 µM SLFN12, a concentration at which the SLFN12 is predicted to be predominantly monomeric (Fig. 7d), and found that SLFN12 did not affect the integrity of the substrate rRNA by itself. Addition of PDE3ACAT resulted in a measurable decrease in rRNA integrity, and further addition of DNMDP or estradiol6, but not trequinsin, greatly enhanced rRNA degradation. Consistent with a requirement for the C-terminal helix of SLFN12 in binding to PDE3A, no activation of SLFN12 RNase activity was observed upon incubation of the isolated SLFN12 NTD with DNMDP-bound PDE3A (Fig. 7e). Taken together, these results support the hypothesis that SLFN12 encodes an RNase, that SLFN12 RNase activity is stimulated by DNMDP-induced complex formation with PDE3A, and that SLFN12 RNase activity is required for DNMDP-induced cancer cell death.


We have performed a comprehensive structure–function analysis of the DNMDP-induced PDE3A-SLFN12 complex that provides significant insight into how the complex forms, the role of DNMDP in stabilizing the complex, and the function of SLFN12 in cellular response to DNMDP. DNMDP binds the active site of PDE3A, forming a surface for high-affinity interaction of SLFN12. Because DNMDP-bound PDE3A forms an adhesive surface for SLFN12, we propose to name this class of compounds “velcrins”. The set of known velcrins currently includes DNMDP, zardaverine, BRD9500, estradiol, anagrelide, nauclefine, and several progesterone receptor agonists1,2,4,5,6,7,26. We propose that velcrins promote dimerization of SLFN12 by recruiting two SLFN12 monomers to a constitutive PDE3A dimer, with the second SLFN12 monomer likely binding cooperatively. We hypothesize that dimerization of SLFN12 may stimulate SLFN12 RNase activity, analogously to activation of RNaseL by 2’–5’-linked oligoadenylates27. Dimerization of SLFN12 likely further stabilizes the PDE3A-SLFN12 complex, as the HDX data identified reduced deuterium uptake at the PDE3A homodimerization interface when SLFN12 was present.

The Cryo-EM structure of the PDE3ACAT-SLFN12 complex revealed that the majority of contacts between the two proteins are made by a single α-helix comprised of 10 amino acids in the C-terminal region of SLFN12 in each monomer. Residues of this C-terminal helix make multiple contacts to PDE3A, including to PDE3A F914, an amino acid determined by DMS to be essential for complex formation and velcrin response. No other human SLFN family members share the primary sequence found in this region of SLFN12, perhaps explaining why no other SLFNs have been found to complex with PDE3A, with or without velcrin treatment. DNMDP provides additional contacts to residues in SLFN12 that stabilize the complex. However, in the case of PDE3A-bound trequinsin, a direct steric clash with residues in the C-terminal region of SLFN12 prevents complex formation. The biophysical and structural studies also implicate electrostatic interactions as important for stability of the PDE3A-SLFN12 complex.

SLFN11, SLFN13, and SLFN14 all encode an RNase activity18,19,20 and have been shown to play a role in restriction of viral infection20,28,29. We speculate that SLFN12 may have a similar physiological function that is leveraged by velcrins to kill cancer cells expressing elevated levels of PDE3A and SLFN12. Velcrins have been shown to synergize with the anti-viral, cell death-inducing interferons4, and another component of the anti-viral innate immune response, RNase L, is able to mediate killing of virally infected cells30,31,32, establishing a precedent for cytotoxic response to RNase activation. Here we demonstrate that SLFN12 is also an RNase, that this RNase activity is increased by incubation with velcrin-bound PDE3A, and that SLFN12 RNase activity is essential for velcrin-induced cancer cell killing.

Unlike traditional targeted therapies that leverage dependencies created by genomic alterations in cancer cells, velcrins instead induce cancer cell death by a novel gain-of-function mechanism mediated by PDE3A-SLFN12 complex formation. Inhibition of PDE3A enzymatic activity does not correlate with cancer cell killing caused by PDE3A-SLFN12 complex formation1 and may not be required26. PDE3A-SLFN12 complex formation is thus the critical event in velcrin response, and the structure reported here provides the molecular details of this interaction. Further development of our understanding of the mechanism of action of the velcrin-induced PDE3A-SLFN12 complex in cancer cell killing will support evaluation of velcrins as potential cancer therapeutics.

Article classification: Biological abstract
Share to:
Add:Room A11-329, 1st Floor, No.1, SBI Venture Street, Optics Valley, East Lake
New Technology Development Zone, Wuhan, China.
Certificate NO.:U18Q28010569R0S