Alterations to the broad-spectrum formin inhibitor SMIFH2 modulate potency but not specificity

SMIFH2 is a small molecule inhibitor of the formin family of cytoskeletal regulators that was originally identified in a screen for suppression of actin polymerization induced by the mouse formin Diaphanous 1 (mDia1). Despite widespread use of this compound, it is unknown whether SMIFH2 inhibits all human formins. Additionally, the nature of protein/inhibitor interactions remains elusive. We assayed SMIFH2 against human formins representing six of the seven mammalian classes and found inhibitory activity against all formins tested. We synthesized a panel of SMIFH2 derivatives and found that, while many alterations disrupt SMIFH2 activity, substitution of an electron-donating methoxy group in place of the bromine along with halogenation of the furan ring increases potency by approximately five-fold. Similar to SMIFH2, the active derivatives are also pan-inhibitors for the formins tested. This result suggests that while potency can be improved, the goal of distinguishing between highly conserved FH2 domains may not be achievable using the SMIFH2 scaffold.


Results and discussion
SMIFH2 was selected from a library of drug-like molecules for its ability to inhibit the mouse formin mDia1 in a pyrene-actin polymerization assay 18 . We used the pyrene-actin polymerization assay to characterize the inhibitory activity against a panel of human formins representing six of the seven classes of mammalian formins (Fig. 1A). These formins were expressed recombinantly as constitutively active 'FFC' fragments (containing FH1, FH2, and C-terminal regions; Fig. 1B) and purified using affinity chromatography (Fig. 1C).
SMIFH2 and its analogs were synthesized as described in the Experimental Procedures. In all NMR characterization experiments, it was observed that SMIFH2 was a mix of E and Z isomers, with a doubling of each proton peak due to the different environments in each isomer. In some cases, an uneven distribution of isomers was observed immediately after dissolving SMIFH2, as shown for the solvent deuterated tetrahydrofuran (THFd 8 ) (Fig. 2). After incubating at room temperature for 20 h, this sample equilibrated to a 1:1 mixture of isomers, as judged by the equivalent peak integrations for each pair. The shift from unequal to equal peak integrations indicates that the equilibrium constant for the isomerization process is close to 1 and that the molecule exchanges between these isomers on an experimentally relevant timescale. From NMR analysis of SMIFH2 in deuterated dimethyl sulfoxide (DMSO-d 8 ), the initial ratio of isomers was closer to 1:1 than was observed in THF-d 8 . Based on this, we infer that the equilibration kinetics are faster in DMSO-d 8 than in THF-d 8 . Regardless of solvent, we cannot say which peak of each pair corresponds to which isomer. Furthermore, it is unknown whether both isomers or just one isomer inhibits the FH2 domain.
All FFC fragments were active in pyrene-actin polymerization assays, as judged by their ability to stimulate assembly over the baseline rate of actin/profilin alone (Fig. 3, comparing red and gray curves). As reported previously, diaphanous related formins (DIAPH1 and DIAPH2) and inverted formin 2 (INF2) stimulated assembly at low nanomolar concentrations (5 nM). FMNL3, FMN2, and DAAM2 had intermediate nucleation efficiency  2 To assay diverse human formins, we selected representatives of six of the seven mammalian formin subclasses. (B) 'FFC' constructs, which include the FH1 and FH2 domains and all C-terminal sequences, were isolated for each formin (DID, Diaphanous Inhibitory Domain; DAD, Diaphanous Autoregulatory Domain). (C) SDS-PAGE showing the purity of each recombinant formin FFC fragment after purification by Coomassie staining (see Figure S7 for uncropped gel). To characterize the specificity of SMIFH2, we titrated the inhibitor against each purified formin, controlling for the volume of DMSO for all polymerization reactions (Fig. 3, as indicated by blue traces). In all cases, SMIFH2 had a concentration for 50% inhibition (IC 50 ) in the 10-20 µM range. This is comparable to the reported value of 15 µM for mouse Dia1 18 . Among the diaphanous formins, SMIFH2 has a more potent effect on DIAPH2 (Dia3) than on DIAPH1 (Dia1), a notable reversal of preference compared to the quinoxaline-based inhibitor reported by Higgs and coworkers (compound 2 in Gauvin et al.) 20 .
Several SMIFH2 analogs are known to retain or lose inhibitory activity against mDia1 18 . We sought to expand this dataset to more extensively characterize structure-activity relationships in the SMIFH2/formin complex. All 17 analogs shown in Fig. 4 were synthesized according to the scheme described in the Experimental Procedures ( Fig. 6) and characterized by proton NMR (Fig. S8). Each inhibitor was tested for potential interaction with actin in the absence of formin (Figs. S3, S4). At the highest concentrations tested, we found that some inhibitors had modest effects on the polymerization of 2 µM actin (Fig. S3). These effects were less apparent in the presence of 4 µM profilin (Fig. S4), indicating that the profilin/actin interaction may mask the inhibitor interaction or suppress polymerization of a potential actin/SMIFH2 complex. This panel of 18 inhibitors was tested against five formin FFC fragments (DIAPH1, DIAPH2, INF2, FMNL3, and FMN2), and the IC 50 values were measured ( Fig. 4 and Table 1). One notable trend in this dataset is that the lack of specificity, noted above for SMIFH2 (Fig. 3), holds for its analogs: no modifications made to SMIFH2 led to a selective inhibitor, at least among the five formins that were tested. In other words, active inhibitors inhibited all formins and inactive inhibitors lacked activity against all formins tested.
Modifications to the thiobarbiturate core were of two types: alkylation or arylation of N3 and replacement of the thiocarbonyl with an oxygen carbonyl (atom numbering in gray in Table 1 structure). The latter substitution (5) led to a complete loss of inhibitory activity ( Fig. 4 and Table 1), consistent with data reported by Kovar and coworkers (compound 3 in Rizvi, et al. 18 ). Both methylation (2) and arylation (4) of the SMIFH2 N3 were well tolerated, and the N3 phenyl analog (4) had 3-fourfold more potent inhibitory activity than SMIFH2. Surprisingly, the ethyl-substituted analog (3) had no detectable activity. This was unexpected since the high activity of Figure 2. SMIFH2 is a mix of isomers that equilibrate over time. 1 H NMR spectra of the same sample were collected at two time points. The blue spectrum (bottom) corresponds to the SMIFH2 sample immediately after preparation in THF-d 8 . The uneven peak integrations are evident in the signals for protons A, B, C, D and E. The initial isomeric ratio was calculated as 3.5:1 by integration of major resonance at 8.25 ppm and minor resonance at 8.33 ppm for proton E. After 20 h at 25 °C (red spectrum; top), the peak integrations were similar for each isomer, reflecting equilibration to a ≈1:1 E:Z mixture (isomeric ratio 1 www.nature.com/scientificreports/ compound 4 showed that the binding pocket on the FH2 domain could accommodate substituents as large as a phenyl ring. One defining feature of PAINs is non-sensical structure-activity relationships 26 , and it is possible that activity data for compounds 2, 3, and 4 are confounded by the other factors apart from the free energy change associated with protein/inhibitor binding. The tolerance of SMIFH2 to methylation (2) and arylation (4) of N3 further supports a model where a carbonyl tautomer (as drawn in Table 1), not a thio enol or oxygen enol 29,30 , is the active form of SMIFH2. The activity of compounds 2 and 4 also indicates that the -NH group of SMIFH2 most likely does not act as a hydrogen bond donor in the protein binding pocket. Meanwhile, several alterations to the furan-ring portion of SMIFH2 had a dramatic effect on activity. Both the pyrrole (6) and thiophene (7) analogs had no detectable inhibitory effect on any formin tested. On the other hand, halogenation of the furan ring at the C5' position, either with Cl (15) or Br (16), was well tolerated, while methylation at that same position (17) abolished activity. This is consistent with the high activity of an analog with another halogen (I) at the 5' position, reported by Kovar and coworkers (compound 2 in Rizvi, et al. 18 ). We synthesized an analog with an extended linker between the furan and central rings. This compound (18) had modest activity.
SMIFH2 has a Br in the meta position of the N1 ring (R 2 position in Table 1). The effect of removing this Br was explored with compound 9, which showed measurable but weak to minimal activity. Compound 9 was poorly soluble in DMSO, and had to be dissolved in DMF for assays. Aside from removing the Br altogether, we found that moving this Br to the para position (8) maintained or slightly enhanced inhibitory activity. We were interested in the effects of a methoxy group on the N1 ring, and we successfully synthesized the para-methoxy analog (10). While compound 10 had reduced activity compared to meta-Br (1) or para-Br (8) SMIFH2, substitution of the furan with either Br or Cl at the C5' position (11 or 12, respectively) or Br at the C4' position (13) in addition to the N1 para-methoxy phenyl group appeared to have synergistic effects. This was most striking with compound 12, the most potent inhibitor in our series of compounds. Including these two substitutions together decreased IC 50 values by two-ninefold (comparing compounds 1 and 12), depending on the formin. For instance, the IC 50 against INF2 was decreased from 10 ± 5 µM for SMIFH2 (1) to 2 ± 1 µM for compound 12. Unlike incorporation of a halogen at the furan C4' or C5' positions in the para-methoxy series (11,12,13), addition of a Cl at furan C5' (14) did not have increased potency compared with para-Br alone (8).
Taken together, the activity data for SMIFH2 and the panel of 17 analogs highlight two essential components of the SMIFH2 structure: the thiocarbonyl and the furan ring. Our results also indicate that SMIFH2 derivatives maintain activity when the structure is modified at the thiobarbiturate N3 position and also at the meta and para positions on the N1-phenyl ring and the 4' and 5' positions of the furan ring. Despite the chemical diversity explored in this study, we did not identify an isoform-specific formin inhibitor. Such an inhibitor may exist in another region of SMIFH2 chemical space not explored here, such as ortho and/or multiple substitutions on the N1-phenyl ring or furans attached to the beta carbon at their 3' carbon, among many other potential modifications. Alternatively, the lack of specificity among the panel of molecules in this study suggests that a more specific inhibitor may require a scaffold completely different from the thiobarbiturate core of SMIFH2 in  To date, no atomic-resolution structural data is available for SMIFH2 bound to the FH2 domain, hindering the rational design of structural perturbations to the inhibitor. This is in contrast to the Arp2/3 inhibitor CK666 22 , whose x-ray crystal structure was used as a starting point for free energy perturbation calculations to optimize inhibitor/protein interactions 31 . The more potent analogs reported here may be candidates for structural studies of a protein/SMIFH2 complex, particularly with the human formin DIAPH2 (ortholog of mDia3). Based on IC 50 values, SMIFH2 analog 12 may have a higher affinity to DIAPH2 than any other formin/inhibitor pairs measured, although this would need to be confirmed by equilibrium experiments.
To our knowledge, it is uncertain where SMIFH2 lies in the taxonomy of covalent/noncovalent or reversible/ irreversible inhibitors. Its α/β unsaturated dicarbonyl motif is likely a Michael acceptor, which would react with nucleophilic amino acids, placing SMIFH2 in the covalent category [32][33][34] . The furan substitution at the β carbon of SMIFH2 suggests that addition of a thiol group could be reversible 35 . Similar compounds to SMIFH2 with a modified thiobarbiturate core have been found to inhibit mushroom tyrosinase irreversibly 36 , either through covalent modification of an active site residue or induction of protein misfolding. Cellular studies with SMIFH2 have shown reversal of SMIFH2's effects after a washout [37][38][39] . However, whether this is due to new protein synthesis or dissociation of the SMIFH2/formin complex is unknown.
Motivated by the wide range of inhibitory activities observed in our structure-activity study ( Fig. 4 and Table 1) and the potential for covalent targeting by SMIFH2, we used computational modeling to explore differences in molecular structure and reactivity among this group of molecules. We used Gaussian density functional theory calculations at the B3LYP/6311G + + (d,p) level of theory and basis set to optimize the geometry of a subset of ten molecules, selected because of their variable activities. Figure 5A shows their optimized geometries in water for both the E and Z isomers. The optimized geometry was not sensitive to basis set or solvent (Fig. S5). www.nature.com/scientificreports/ Overall, the molecular shape was similar for all molecules, regardless of their inhibitory activity. Torsion angles for 'peripheral' rings with respect to the central thiobarbiturate ring were between 89.9° and 91.5° for the bond connecting N1 of the central ring to the equivalent of the Br-phenyl ring of SMIFH2 and between 179.5° and 180° for the bond connecting C β of the α/β-unsaturated thiobarbiturate to C2' of the furan ring. We probed the degree of reactivity of SMIFH2 and these nine derivatives using a global electrophilicity index 40 , which we calculated from single point energies of water-solvated neutral, anionic and cationic species. We found a trend of higher electrophilicity for active inhibitors (dark bars in Fig. 5B). However, at least one inactive molecule (7), which has a thiophene in place of the furan, has an electrophilicity value slightly larger than SMIFH2 (Fig. 5B). This indicates that electrophilicity may be a determinant, but not the sole determinant, of activity. Consistent with higher electrophilicity, active inhibitors also had slightly smaller HOMO/LUMO energy gaps (Fig. 5C). Finally, we calculated Hirshfeld charges for carbons that were potential sites of nucleophilic attack, namely the carbonyl carbons and the carbon in the β position relative to the carbonyls. These charges were not correlated with activity (Fig. 5D). Overall, our computational results indicate that these SMIFH2 derivatives have similar geometries and variation in electrophilicity.
The widespread use of SMIFH2 in the decade since the foundational work of Kovar and coworkers 18 underscores the demand for small molecule formin inhibitors. In addition to the search for formin-specific inhibitors, recent data showing activity of the formin inhibitor SMIFH2 on myosins 25 have motivated a search for formin inhibitors with minimal off-target interactions. It will be interesting to assay the series of molecules from this study for myosin inhibition and to determine whether potency against myosins and potency against formins are correlated. Moving forward, a deeper understanding of mechanisms and specificity of formin-targeting small-molecule inhibitors is important for understanding the roles of formins in cells. We found that while many alterations disrupt SMIFH2 activity, we could increase potency by approximately fivefold. However, we failed to increase specificity, suggesting that while potency can be improved, achieving specificity among closely related FH2 domains may be hard to achieve with the SMIFH2 scaffold.   DAAM2-FFC-expressing cells were lysed, centrifuged, and passed over a HisTrap column as described above for INF2. Pure fractions were pooled and buffer exchanged by PD10 column (Cytiva) into SEC buffer before dialyzing with 1:1 glycerol:SEC buffer overnight. Aliquots were flash frozen and stored at -80 °C.
DIAPH2-FFC expressing cells were lysed by French Press in Ni-NTA lysis buffer supplemented with PMSF, DNaseI and protease inhibitor cocktail and centrifuged, as described for INF2. Clarified lysates were nutated for 1 h at 4 °C with Ni-NTA resin (Thermo Sci). The resin was washed with 25 column volumes (CV) of Ni-NTA lysis buffer and 25 CV of Ni-NTA lysis buffer supplemented with 10 mM imidazole, before eluting with lysis buffer supplemented with 500 mM imidazole. Eluted protein was buffer exchanged by PD10 column into S-buffer (10 mM NaCl, 10 mM PIPES pH 6.5, 1 mM EDTA, 1 mM DTT). The protein was loaded onto a 1-mL SP-FF cation exchange column (Cytiva) and eluted with a gradient from 10 to 500 mM NaCl over 30 CV. Selected fractions were exchanged into SEC buffer by PD10 column and then dialyzed into 1:1 glycerol:SEC buffer overnight. Aliquots were flash frozen and stored at -80 °C. FMN2-FFC-expressing cells were lysed in a modified pH 6.5 Ni-NTA lysis buffer. The rest of the purification protocol was identical to that described for DIAPH2-FFC.
FMNL3-FFC-expressing cells were lysed in TALON extraction buffer (300 mM NaCl, 50 mM NaPi pH 8.0, 1 mM β-mercaptoethanol (βME)) supplemented with 1 mM PMSF and 4 µg/mL DNaseI. After lysis and centrifugation as described above for INF2-FFC, the supernatant was nutated with 1 mL TALON resin (Takara) for 30 min at 4 °C. The resin was washed with 25 CV TALON extraction buffer and then 25 CV TALON wash buffer (300 mM NaCl, 50 mM NaPi pH 7, 1 mM βME) before eluting with TALON wash buffer supplemented with 200 mM imidazole. The eluted protein was dialyzed against 50 mM NaPi pH 7, 50 mM NaCl, 1 mM DTT and loaded onto an SP-FF cation exchange column and eluted with a gradient from 50 to 500 mM NaCl over 30 CV. Fractions were dialyzed against S-buffer, loaded on an SP-FF column, and eluted with a step gradient from 10 mM NaCl to 500 mM NaCl. Eluted fractions were loaded onto a Superdex200 10/300 column (Cytiva) equilibrated with SEC buffer. Pure fractions were flash frozen in liquid nitrogen and stored at -80 °C. FMNL3 was also purified in a protocol that skipped the first SP-FF column, and there was no difference in activity or purity. DIAPH1-FFC-expressing cells were lysed by probe sonicator in TALON extraction buffer supplemented with PMSF and DNaseI. The formin was purified using TALON resin as described above for FMNL3-FFC. Eluted fractions were dialyzed against Q-buffer (10 mM Tris pH 8, 1 mM DTT) overnight with 100 units of thrombin to cleave the C-terminal histidine tag. The cleaved protein was loaded onto a monoQ anion exchange column (Cytiva) and eluted with a gradient of 0 to 500 mM KCl over 30 CV. Semi-pure fractions were buffer exchanged into TALON extraction buffer using a PD10 column and incubated with TALON resin to remove uncleaved protein. The fraction that did not bind TALON was collected and dialyzed against Q-buffer and then 1:1 glycerol:Q-buffer. This results in a version of DIAPH1-FFC that is truncated by ~ 10-20 residues on the C-terminus as assessed by MALDI mass spectrometry (data not shown).
Delphilin-FFC was purified as described for the human FFC isoform 28 . Profilin from Schizosaccharomyces pombe was expressed in a pET plasmid in BL21-DE3* cells and was purified as described 41 . We chose this isoform of profilin since, compared to other profilins, its interaction with actin is less sensitive to actin modification at Cys734 42 . Actin was purified from rabbit skeletal muscle and labeled with pyrene-iodoacetamide as described 42 . For pyrene-labeled actin, the pyrene concentration was calculated using an extinction coefficient of 21,978 M −1 cm −1 at 344 nm, and the actin concentration was calculated using the correction factor of 0. Pyrene-actin assays. Assays were carried out as described 47 . Rabbit skeletal muscle actin (5% pyrene labeled) was incubated for 2 min at 25 °C with 200 μM EGTA and 50 μM MgCl 2 to convert Ca-actin to Mg-actin. When included in the experiment, a 2:1 molar ratio of profilin was incubated with actin for 2 min at 25 °C before conversion to Mg-actin. Polymerization was initiated by adding polymerization buffer (KMEH, final concentration: 10 mM HEPES, pH 7.0, 1 mM EGTA, 50 mM KCl, 1 mM MgCl 2 ). Prior to mixing KMEH with actin, formins and then the inhibitor in DMSO or DMF was added at ≤ 1% of the total volume. Finally, this mixture was rapidly mixed with Mg-actin, with approximately 10-20 s dead time prior to measurement. Fluorescence was monitored every 15 s using a TECAN F200 Pro with λ ex = 360 ± 17 nm and λ em = 415 ± 10 nm. The same volume of organic solvent (DMSO or DMF) was added to each well regardless of inhibitor concentration.
Computational analysis. SMIFH2 isomers and their chemical analogs were constructed using the program Avogadro 48 . All quantum mechanical calculations were carried out using Gaussian 16 (versions A-2016 and C-2019) 49 . Geometry optimizations, single point energy, and Hirshfeld population calculations were performed with the unrestricted (U) B3LYP functional with the 6-311 + G(d,p) basis set [50][51][52][53][54][55][56][57][58][59][60][61][62] . For geometry optimizations, the opt keyword was set to the 'tight' option to ensure adequate convergence. The initial optimization for all compounds was performed in vacuum and used further for solvated calculations reported in this study. Single point energy calculations, for a global electrophilicity analysis, at the neutral, anionic, and cationic states were performed with the unrestricted (U) PW6B95 functional supplemented with Grimme et al.'s D3 empirical diffuse function and the aug-cc-pVTZ basis set [63][64][65][66][67] . SCRF polarizable continuum model (PCM) was used to model an implicit water as the solvent. Optimized geometries were visualized using PyMOL 68 .
To calculate global electrophilicity (ω), we used the convention of Parr et al. ω = µ 2 /(2η) where µ and η represent the electronic chemical potential and the chemical hardness, respectively 40,69-72 . Ionization energy and electron affinity were used to calculate µ and η, as defined by De Vleeschouwer et al. 73 We used the (U) PW6B95-D3 functional and aug-cc-pVTZ basis set as described above to determine the global electrophilicity of SMIFH2 and its analogs more accurately 74 .

Data availability
Characterization data for compounds synthesized in this study are included in the Supporting Information. The data used to calculate averages in Table 1 are listed in Table S1. Additional datasets generated during the current study are available from the corresponding author on reasonable request.