Rational development of Stafib-2: a selective, nanomolar inhibitor of the transcription factor STAT5b

The transcription factor STAT5b is a target for tumour therapy. We recently reported catechol bisphosphate and derivatives such as Stafib-1 as the first selective inhibitors of the STAT5b SH2 domain. Here, we demonstrate STAT5b binding of catechol bisphosphate by solid-state nuclear magnetic resonance, and report on rational optimization of Stafib-1 (Ki = 44 nM) to Stafib-2 (Ki = 9 nM). The binding site of Stafib-2 was validated using combined isothermal titration calorimetry (ITC) and protein point mutant analysis, representing the first time that functional comparison of wild-type versus mutant protein by ITC has been used to characterize the binding site of a small-molecule ligand of a STAT protein with amino acid resolution. The prodrug Pomstafib-2 selectively inhibits tyrosine phosphorylation of STAT5b in human leukaemia cells and induces apoptosis in a STAT5-dependent manner. We propose Pomstafib-2, which currently represents the most active, selective inhibitor of STAT5b activation available, as a chemical tool for addressing the fundamental question of which roles the different STAT5 proteins play in various cell processes.

Transcription factors orchestrate cellular signalling by regulating transcription of their target genes, thus allowing precise regulation of cellular phenotype 1 . They do not possess enzymatic activities, making their functional manipulation with cell-permeable small molecules more challenging. The transcription factors STAT5a and STAT5b in particular are highly homologous 2 and are frequently referred to jointly as "STAT5", implying that they carry out identical functions. However, while some protein functions are indeed redundant, others are not. For example, although both STAT5a and STAT5b are constitutively activated in numerous human cancers, including human leukaemias harbouring the Philadelphia chromosome 3 which leads to expression of the Bcr-Abl fusion protein, the inhibition of STAT5b was shown to reduce tumour cell proliferation more than the inhibition of STAT5a did 4,5 . Small-molecule inhibitors which differentiate between the two STAT5 proteins would be highly beneficial for clarifying their individual roles. The most effective and selective approach by which to inhibit STAT proteins involves functional inhibition of the protein-protein interaction domain, the Src homology 2 (SH2) domain 6,7 . However, for most STAT5 inhibitors developed to date, including chromone-based compounds 8,9 , fosfosal 10 , salicylic acid-based STAT5 inhibitors 11,12 , an adenosine-5′-monophosphate derivative 13 , and an osmium complex 14 , selectivity for one STAT5 protein over the other was either minimal or not reported. We recently presented catechol bisphosphate (1, Fig. 1a) and its derivatives Stafib-1 (2, Table 1) 15 and Capstafin 16 as selective inhibitors of the STAT5b SH2 domain.
Despite overwhelming evidence for the biological significance of STAT proteins as therapeutic targets, the development of small-molecule STAT inhibitors has been hampered by the absence of structural characterization of complexes between STATs and small molecules by X-ray crystallography or NMR spectroscopy. The isolated STAT SH2 domains are mostly insoluble; virtually all research studies to date have used STAT constructs containing multiple STAT domains 17,18 . The molecular weight of the proteins expressed by these constructs is too high for deducing detailed structural information of the STAT-ligand complex by solution-state NMR methods.
We have employed solid-state NMR spectroscopy to verify and characterize the binding of catechol bisphosphate to STAT5b, an interaction which provides the foundation of our docking-based model for the binding of Stafib-1 to STAT5b 15 . Following structure-guided optimization of Stafib-1, we performed a comparative analysis of binding of the optimized inhibitor to wild-type STAT5b protein versus binding to mutant STAT5b proteins using isothermal titration calorimetry (ITC). Here we present the design and binding mode validation of the optimized STAT5b inhibitor Stafib-2, and its prodrug Pomstafib-2, for use as a research tool in cell-based assays.

Results and Discussion
Characterization of catechol bisphosphate binding to STAT5 by solid-state NMR. Since binding of 1 to STAT5b Arg618 in the SH2 domain had been indicated in a fluorescence polarization (FP)-based assay 15,19 , we decided to investigate the mode of STAT5b-binding of 1 using solid-state NMR spectroscopy. The use of solid-state NMR for the characterization of binding between a small molecule and an SH2 domain has not previously been reported. 13 C direct polarization (DP) / magic-angle spinning (MAS) spectra of 13 C 6 -1 revealed that the three 13 C resonances of 13 C 6 -1 split into sets of two in the presence of STAT5b (Fig. 1b). Peak assignment for unbound 1 was carried out on the basis of 2D 13 C-13 C correlation spectra of 13 C 6 -1 ( Supplementary Fig. S1). Similarly, the single 31 P DP/MAS-NMR resonance of 13 C 6 -1 in the absence of STAT5b split into two peaks in the presence of protein, indicative of binding (Fig. 1c). Moreover, the signal width of the two 31 P-NMR resonance signals was unequal (full widths at half maximum of 94.4 Hz and 147.0 Hz, respectively, Fig. 1c inset). This suggests differences in the rigidity of the chemical environment between the two phosphate groups, indicating that one or more additional amino acids beside Arg618 must contribute to binding of 1. Inspection of a homology model of the STAT5b SH2 domain 15 for further putative interaction partners for the phosphate groups suggested Lys600 (Fig. 1d). In fact, a STAT5b Lys600Ala mutant bound with significantly weaker affinity than wild-type STAT5b 15 to a fluorophore-labelled derivative of 1 in an FP-based assay (Fig. 1e), indicating that STAT5b Lys600 interacts with at least one of the two phosphate groups of 1. We cannot exclude the possibility that STAT5b amino acid residues other than Arg618 and Lys600 are also involved in binding to 1.
Structure-guided optimization of Stafib-1. After confirming binding of 1 to the phosphotyrosine binding site of the STAT5b SH2 domain, we aimed to exploit the docking-based binding pose for structure-guided optimization of the catechol bisphosphate derivative Stafib-1 (2, Table 1). The terminal phenyl ring of Stafib-1  13 C DP/MAS-NMR of 13 C 6 -1 in buffer in the absence (black) and presence (red) of STAT5b. The spectrum of non-isotopically enriched 1 in the presence of STAT5b is shown in blue. (c) 31 P DP/MAS-NMR of 13 C 6 -1 in the absence (black) and presence (red) of STAT5b (recycle delay: 2.5 s) Inset: a pure Lorentzian function was applied to fit the experimental spectrum of 13 C 6 -1 in the presence of STAT5b (recycle delay: 15 s). Deconvolution produced two fits (sites I and II) of STAT5b-bound 13 C 6 -1 with the equal relative integral areas. The sum of the Lorentzian fits is shown as dotted curve. (d) Binding mode of 1 to the STAT5b SH2 domain as predicted by AutoDock Vina 15 . The figure was generated using PyMol 37 . (e) Binding between a fluorophore-labelled derivative of 1 to STAT5b wild-type (previously published in) 15 or the STAT5b Lys600Ala mutant analysed by fluorescence polarization. Error bars represent standard deviations from three independent experiments, except for STAT5b Lys600Ala at 2.56 µM (n = 2). is predicted to reside in a hydrophobic pocket delineated by the side chains of Phe633 and Tyr665 ( Fig. 2a and Supplementary Fig. S2). The distance and the angle between these two aromatic side chains suggested to us that this pocket would be ideally occupied by two aromatic rings connected via an sp 3 -hybridized centre. An obvious chemical moiety is the phenoxyphenyl group, which can be generated by adding a p-phenoxy substituent to the terminal phenyl ring of 2. While the p-phenoxy-substituted Stafib-1 derivative 3 was 2-fold less active than 2 (K i = 0.093 ± 0.003 µM for 3 vs. 0.044 ± 0.001 µM for 2, Table 1), we attributed the reduced activity to inappropriate positioning of the added phenoxy group owing to excessive space requirements of 3 in the protein binding pocket. To facilitate correct placement of the phenoxyphenyl group in its putative binding pocket, we replaced the central naphthyl group of 3 with a phenyl group, resulting in compound 4 ( Table 1). Docking of 4 into the STAT5b SH2 domain with AutoDock Vina 20 suggested a tight fit of the phenoxyphenyl group in the binding pocket delineated by Phe633 and Tyr665, along with face-to-face π-stacking interactions between the side chains of the aromatic amino acids and the phenoxyphenyl group (Fig. 2b). 4 displayed a K i value of 0.0088 ± 0.0003 µM, which represents a more than 10-fold reduction in K i value relative to 3, and a 5-fold reduced K i value relative to 2. 4 was found to be highly selective for STAT5b over STAT5a (K i = 1.3 ± 0.1 µM) and other STAT family proteins, as well as the more distantly related SH2 domain of the tyrosine kinase Lck 21 (Fig. 2c, Table 1). Synthesis of 4 (dubbed Stafib-2) was carried out in 8 synthetic steps in an overall yield of 32% (Fig. 2d). Deletion of the terminal phenoxy group (compound 5) or the central phenyl amide group (compound 6) strongly reduced activity (Table 1), consistent with specific protein recognition by 4. Although 4 has an extremely high affinity for STAT5b, this affinity is crucially dependent on the bisphosphorylated catechol core, as indicated by the lack of activity of the unphosphorylated derivative 7 ( Table 1).

Validation of the docking-based binding mode of Stafib-2 to STAT5b by ITC. In order to
verify that the pocket delineated by Phe633 and Tyr665 is indeed involved in binding 4, we investigated the protein-ligand interaction using isothermal titration calorimetry (ITC). Injection of 4 into a solution of wild-type STAT5b revealed tight binding (K d = 0.32 ± 0.11 µM) with both enthalpic (ΔH = −4.1 ± 0.5 kcal mol −1 ) and entropic contributions (−TΔS = −4.8 ± 0.6 kcal mol −1 ), consistent with binding mediated by both electrostatic   and hydrophobic interactions (Fig. 2e,f) 22 . In contrast, binding of 4 to a STAT5b Phe633Ala, Tyr665Ala double mutant was over tenfold weaker (K d = 3.8 ± 1.7 µM, Fig. 2e,g). Of note, while the change in enthalpy was comparable (ΔH = −5.1 ± 0.5 kcal mol −1 ), the entropic contribution of double-mutant binding (− TΔS = −2.4 ± 0.3 kcal mol −1 ) is only half as strong as it is for binding of 4 to wild-type STAT5b. Alanine mutation of STAT5b Phe633 alone was less detrimental for binding (K d = 1.65 ± 0.02 µM, Fig. 2e,h), with a virtually identical change in enthalpy (ΔH = −4.3 ± 0.2 kcal mol −1 ), but a lesser gain in entropy (−TΔS = −3.6 ± 0.2 kcal mol −1 ) as compared to wild-type. These data confirm that both Phe633 and Tyr665 are involved in the interaction. Mutation of Arg618 (K d = 5.6 ± 1.7 µM, Supplementary Fig. S3) or Lys600 (K d = 23 ± 10 µM, Supplementary  Fig. S3) to alanine also reduced binding affinity, consisted with the docking pose (Fig. 2b). Binding of 4 to STAT5a demonstrated an approximately 20-fold lower affinity (K d = 6.1 ± 3.2 µM, Supplementary Fig. S3), confirming the specificity of 4 for STAT5b observed in the competitive binding assays. This represents the first time that the binding site of a small-molecule ligand of a STAT protein has been characterized with amino acid resolution by functional comparison of wild-type versus mutant proteins using ITC. Of note, studies carried out on the close homologue STAT5a have shown the hydrophobic pocket delineated by Phe633 and Tyr665 to be important for STAT5a activity 23 , consistent with the likelihood of a functional relevance for the corresponding hydrophobic Stafib-2 binding pocket of STAT5b.
Validation of Pomstafib-2 in cell-based assays. Human K562 leukaemia cells harbour the tyrosine kinase Bcr-Abl, which leads to constitutive activation of STAT5 (Fig. 3a). In order to test whether the selective activity of Stafib-2 (4) also extends to STAT5b in living cells, we masked its negative charges by conversion to the pivaloyloxymethyl ester 8, which was dubbed Pomstafib-2 (Fig. 3b). Pivaloyloxymethylesters have been successfully shown to mediate cell-permeability of organic phosphates and phosphonates [24][25][26][27][28] . In cells, the pivaloyloxymethyl groups of 8 are cleaved off by intracellular esterases, thereby releasing the parent compound, along with formaldehyde and pivalic acid. Because commercially available antibodies do not allow for distinction between STAT5a phosphorylated at Tyr694 and STAT5b phosphorylated at Tyr699, we used an assay system based on fusion proteins of the individual STAT5 proteins with GFP, which are also recognized by anti-phospho-STAT5 antibodies, but can be distinguished from endogenous STAT5 by their higher molecular weight 15,16 . Treatment of STAT5b-GFP-transfected K562 cells, in which STAT5 proteins are constitutively activated by Bcr-Abl, with 8 resulted in a dose-dependent decrease in phosphorylation at STAT5b Tyr699, with an IC 50 of 1.5 µM (Fig. 3c,d).
In contrast, phosphorylation of STAT5a at Tyr694 was not significantly reduced by 8 in STAT5a-GFP transfected cells (Fig. 3e,f). Activity of 8 against STAT5b phosphorylation is observed as early as 1 h after addition to the cells, with maximum activity between 2 h and 4 h, and is still significant after 8 h of exposure ( Supplementary Fig. S4). 8 is 2.5-fold more active against STAT5b phosphorylation than the pivaloyloxymethylesters of 2 (IC 50 = 3.8 µM) 15 and of Capstafin 16 (IC 50 = 4.1 µM) (Supplementary Fig. S5). The lack of inhibition of STAT5a/b phosphorylation by the pivaloyloxymethyl ester 9 (Fig. 3c-f), which is based on a phosphonate previously shown to be inactive against STAT5a/b 15 , indicates that neither the formaldehyde nor the pivalic acid released during prodrug cleavage contribute to the inhibitory effect. The unmasked bisphosphate 4 did not show significant activity in cells (Fig. 3c-f), indicating poor cell permeability. Phosphorylation of endogenous STAT5 in the STAT5a/b-GFP transfected cells was also inhibited by 8, but to a lesser extent than transfected STAT5b-GFP, which presumably reflects the relative abundances of STAT5b and STAT5a in the cells (Supplementary Fig. S6). Consistent with previous observations 15 , these data imply that the majority of phosphorylated endogenous STAT5 protein in K562 cells is STAT5b. Inhibition of signalling via STAT5b in K562 cells has previously been shown to reduce cell viability and induce apoptosis (Fig. 3a) 4,5,29 . Treatment of untransfected K562 cells with 8 increased the apoptotic rate of K562 cells in a dose-dependent manner (Fig. 3g, Supplementary Fig. S7). To verify that the effects of 8 were mediated by inhibition of STAT5 proteins, we carried out the same experiment with MDA-MB-231 cells, which do not harbour constitutively activated STAT5 16 and therefore can be assumed to grow STAT5-independently. The apoptotic rate of MDA-MB-231 cells was not increased by 8 (Fig. 3h, Supplementary Fig. S8), consistent with the effects on K562 cells being mediated by inhibition of STAT5 proteins. The observation that higher concentrations of 8 are required for induction of apoptosis (Fig. 3g) than for the inhibition of STAT5b phosphorylation (Fig. 3c,d) in K562 cells may reflect the high selectivity of 8 for STAT5b over STAT5a. We hypothesize that uninhibited endogenous, phosphorylated STAT5a could prevent the induction of apoptosis at concentrations of 8 that are already sufficient for selective inhibition of STAT5b phosphorylation.

Conclusions
We report the rational optimization of the STAT5b SH2 domain inhibitor Stafib-1 (2) to Stafib-2 (4), which displays significantly increased activity whilst maintaining high selectivity over the closely related SH2 domain of STAT5a. We provide the first application of solid-state NMR for the analysis of ligand binding to an SH2 domain. The high affinity of Stafib-2 for STAT5b allowed us to perform the first comparative analysis of binding between a small-molecule STAT SH2 domain ligand and wild-type versus mutant STAT proteins by ITC, providing experimental validation of the hydrophilic and hydrophobic Stafib-2 binding pockets of STAT5b. The pivaloyloxymethylester Pomstafib-2 (8) inhibits STAT5b phosphorylation in cultured human leukaemia cells with an IC 50 of only 1.5 µM, without significantly affecting the phosphorylation of STAT5a, and increases the apoptotic rate of human leukaemia cells in a STAT5-dependent manner. Since Pomstafib-2 (8) currently represents the most active, selective inhibitor of STAT5b activation, we propose its use as a chemical tool to dissect the overlapping and non-redundant functions of the two STAT5 proteins in living cells.
Apoptosis assay. Apoptosis assays were performed as previously described 16 . In brief, K562 cells (2.5 × 10 5 cells per well) or MDA-MB-231 cells (1 × 10 5 cells per well) were seeded in 24-well tissue culture plates (Corning #3526), and treated with compound 8 at the indicated concentrations (final DMSO concentration: 0.2%) for 48 h. Cells were harvested after 48 h. MDA-MB-231 cells were washed twice with warm phosphate buffered saline (PBS), followed by incubation with Accutase (BD Bioscience) at 37 °C for 10 min. Neutralization of Accutase and cell resuspension was carried out with the cell culture supernatant from each well. After harvesting, cells were centrifuged at 3000 rpm at 4 °C for 5 min, washed twice with cold PBS, and centrifuged again. Cells were stained using the PE Annexin V Apoptosis Detection Kit I (BD Bioscience). Cells were resuspended in binding buffer and incubated with PE Annexin V and 7-AAD at 4 °C for 30 min. Apoptosis was measured using a LSR II flow cytometer (BD Bioscience).