Stafia‐1: a STAT5a‐Selective Inhibitor Developed via Docking‐Based Screening of in Silico O‐Phosphorylated Fragments

Abstract We present a new approach for the identification of inhibitors of phosphorylation‐dependent protein–protein interaction domains, in which phenolic fragments are adapted by in silico O‐phosphorylation before docking‐based screening. From a database of 10 369 180 compounds, we identified 85 021 natural product‐derived phenolic fragments, which were virtually O‐phosphorylated and screened for in silico binding to the STAT3 SH2 domain. Nine screening hits were then synthesized, eight of which showed a degree of in vitro inhibition of STAT3. After analysis of its selectivity profile, the most potent inhibitor was then developed to Stafia‐1, the first small molecule shown to preferentially inhibit the STAT family member STAT5a over the close homologue STAT5b. A phosphonate prodrug based on Stafia‐1 inhibited STAT5a with selectivity over STAT5b in human leukemia cells, providing the first demonstration of selective in vitro and intracellular inhibition of STAT5a by a small‐molecule inhibitor.

Protein-protein interactions mediate most biological processes, and their functional modulation by smallm olecules offers vast opportunities for basic research and drug development. [1] However,p rotein-protein interactions represent challenging targets for small molecules, and design approaches for inhibitor development are rare. [2] Phosphorylation-dependent protein-protein interactions are mediated by the phosphorylated side chainso ft yrosine, serine, and threonine residues, and play an important role in signal transduction. We recently proposedO -phosphorylation of preselected natural products as an approachf or the development of non-peptidic and non-reactivel igandso fp hosphorylation-dependent protein-protein interactions. [3] We used this approacht od evelopc atecholb isphosphates [4] as the first chemicale ntitiest hat inhibit the phosphotyrosine-dependent Src homology 2( SH2) domain of the transcription factor STAT5b with high selectivity over the close homologue STAT5a. [5] Both STAT5p roteins are constitutivelya ctivated in numeroush uman tumors. [6] Selective inhibition of either STAT5 protein is desirable for the functional analysiso ft he non-redundantf unctions of STAT5a and STAT5b, [7] andw ould offer flexibility in tailoring the antitumor treatments trategy to individualh uman tumors. Smallm olecule STAT5a inhibitors with selectivity over STAT5b could also serve as therapeutic modalities for age-related osteoporosis. [8] However,n oS TAT5a inhibitors [3,9] with selectivity over STAT5b have been disclosed to date.
Here, we present virtual (in silico) O-phosphorylation of preselected phenolic fragments of natural products, [10] followed by docking-based virtual screening, as an ovel methodology for the identification of inhibitors of phosphotyrosine-dependent protein-protein interaction domains. The initial virtual compound library wasd ownloaded from the ZINC database [11] as ac ollection of 10 369 180 structures. Filtering this database for structural elements described by the structural classification of natural products (SCONP)t ree [10] identified 799 335 compounds ( Figure 1A,s tep 1, Figure S1, and SupportingM ethods in the Supporting Information). Further filtering for fragments with ap henol moiety and am olecular weight below 500 gMol À1 ,a nd removal of certainr eactive moieties (Figure 1A,s tep 2, and Supporting Methods), narrowed down the selection to 85 021 compounds, which were then virtually Ophosphorylated on their phenolic moiety by altering their SMILESs tring ( Figure 1A,s tep 3). [12] Virtual screening of the Ophosphorylated compounds against the STAT3S H2 domain (PBD ID:1 BG1) [13] with AutoDockV ina [14] resulted in 1114 compounds, whichf ulfilledp redefined criteria for the distances betweent he phosphate groupso ft he molecules and the crucial STAT3S H2 domain residues Arg609a nd Lys591 ( Figure 1A, step 4, andF igure S2). [13] After visual inspection of the binding poses, 9molecules (1-9)w ere selected ( Figure 1A,s tep 5, Ta ble S1), which display av ariable degree of resemblance to natural products, depending on the size of the underlying natural product-derived structural element from the SCONP tree. [10] Molecules 1-9 were synthesized by O-phosphorylation of commercially availableo rpre-synthesized phenolic precursors, by at wo-step phosphorylation/debenzylationp rocess ( Figure 1A,s tep 6, Table S1, and Supporting Information), and testedi nafluorescencep olarization( FP) assay against the STAT3S H2 domain ( Figure 1A,s tep 7). [15] Eight of the O-phos-phorylated molecules 1-9 showed ad egree of STAT3i nhibition, with 1-3 inhibiting STAT3b ymore than 40 %a t1 00 mm (Table S1). 1 docked into the STAT3S H2 domain with its phosphate group in close proximity to Arg609a nd Lys591 (Figure 1B). Although screening had been performed with the aim of identifying inhibitors of the STAT3S H2 domain, analysiso f specificity profiles revealed that severalc ompounds, including 1,w ere more active against STAT5a and STAT5b [16] than against STAT3( Ta ble S1). This suggests that the docking approach may not be sensitivee nough to clearly discriminate betweent he STAT family members. Since selectiveS TAT5a inhibitors have not yet been reported,w ed ecided to optimize 1 for binding to STAT5a, rather than STAT3. Compound 1 wasc hosen as a startingp oint for inhibitor development due to its lack of reactive functional groups,a nd its m-terphenyl scaffold should allow forf lexible modifications through Suzuki coupling reactions.
The SH2 domains of STAT5a and STAT5b are 93 %i dentical on the amino acid level, with only 6of91amino acids differing   Figure S3). To investigate the molecular origin of the specificity of 20 for STAT5a over STAT5b, we used wild-type and point mutant STAT5p roteins in FP assays.T he activity of 20 against the point mutant STAT5b Gln636Pro/Met639Asn/Phe640Leu/ Met644Lys/Asn664Ser/Tyr679Phe( dubbed STAT5b-6M), [18] in which the six amino acids of the STAT5b SH2 domain which differ from those in the STAT5a SH2 domain have been replaced by their STAT5a counterparts ( Figure S3), was only marginally increased (46 AE 8% inhibition at 200 mm)c ompared to wild-type STAT5b (37 AE 5% inhibition at2 00 mm), and thus approximately 10-fold lower than the activity against wild-type STAT5a (K i = 10.9 AE 1.8 mm,I C 50 = 22.2 AE 3.6 mm,F igure 3A and Ta ble S3). This indicated that factorso utside the SH2 domain must playasignificant role forbinding of 20 to STAT5a.
We recently described the divergenta mino acids in position 566 of the STAT5l inker domain (Trp in STAT5a, Arg in STAT5b, Figure S3), adjacent to the SH2 domain, as the crucial determinant for selective STAT5b inhibition by small molecules. [18,19] To investigate the role of STAT5a Trp566, we tested 20 against the crossoverp ointm utant STAT5a Trp566Arg, and found its activity to be reduced by 7-fold (K i = 74.0 AE 7.0 mm,F igure3Aa nd Ta ble S3)a sc ompared to wild-type STAT5a. Although this suggests arole for Trp566 in bindingt o20,itisnot the sole determinanto fs pecificity,s ince the reverse mutant STAT5b Arg566Trp (43 AE 7% inhibition at 200 mm)w as only marginally more inhibited than wild-type STAT5b (37 AE 5% inhibition at 200 mm)b y20 ( Figure 3A and Ta ble S3). However,t he combined presentation of the STAT5a SH2 domain and Trp566 in the contexto fS TAT5b, as represented by the mutant STAT5b Arg566Trp/Gln636Pro/Met639Asn/Phe640Leu/Met644Lys/ Asn664Ser/Tyr679Phe (dubbed STAT5b-7M), [18] almost restored binding (K i = 21.1 AE 4.4 mm,F igure 3A and Ta ble S3). These data indicatet hat recognition of 20 by STAT5a depends on both the SH2 domain and Trp566.T he remaining twofold activity difference of 20 between STAT5a (K i = 10.9 AE 1.8 mm)a nd STAT5b-7M( K i = 21.1 AE 4.4 mm,F igure 3A and Ta ble S3) may be mediated by allosteric cross-communication with divergent amino acid positions in the linker domain, [20] the DNA binding domain, [21] or the coiled-coil domain. [22] Docking of 20 into the crystal structureo fS TAT5a [23] using AutoDock FR, [24] with the side chains of amino acids in the phosphotyrosine binding pocket or its immediate vicinity (Trp566,L ys600, Arg618, Asn639, and Lys644)d efined as flexible, placed one of the terminal phenyl rings of 20 near STAT5a Trp566 ( Figure 3B), consistent with p-stacking interactions and with hydrogen bonding between the protona ttachedt ot he indole nitrogen of Trp566 and at least one of the methoxy groups of 20.T he phosphate group of 20 is predicted to form electrostatic interactions with Lys600 and Arg618, which are identicali nS TAT5a and STAT5b, in am anner similart ot hat of phosphotyrosine-containing ligands. [13] Selectivity-conferring electrostatic interactions may arise from interaction with the side chain of STAT5a Lys644,w hich is replaced by methionine in STAT5b. The relevance of STAT5a Lys644 for binding of 20 was demonstrated by reduced inhibition of the crossover double mutant STAT5a Trp566Arg/Lys644Met (44 AE 4% inhibition at 200 mm)a sc ompared to STAT5a Trp566Arg (K i = 74.0 AE 7.0 mm,F igure 3A and Ta ble S3). Conversely,t he activity of 20 against the crossover mutant STAT5b Arg566Trp/Met644Lys was enhanced (K i = 63.1 AE 4.2 mm)a sc ompared to STAT5b Arg566Trp (43 AE 7% inhibition at 200 mm,F igure 3A and Ta ble S3). The triple crossover mutant STAT5b Arg566Trp/ Met639Asn/Met644Lys (dubbed STAT5b-3M, K i = 49.0 AE 5.4 mm, Figure 3A and Table S3) was somewhat more susceptiblet oi nhibition by 20 than the doublem utant STAT5b Arg566Trp/ Met644Lys, suggesting as mall contribution of Asn639( present in STAT5a) to binding.T he remaining 2.5-fold activity gap between the triple mutant STAT5b-3Ma nd the 7-fold mutant STAT5b-7M(K i = 21.1 AE 4.4 mm)p oints towards an allostericcontribution by one or more of the remaining four STAT5a/b SH2 domain divergent amino acids, which are furtherr emoved from the putative binding site of 20.
We note that in the available crystal structure of STAT5a, [23] the position of the side chain of Lys600 is not suitable for phosphate group binding ( Figure S4). Lys600 is required for binding of phosphotyrosine in peptidic SH2 domain ligands, and the unsuitable orientation is presumably ac onsequence of the fact that this STAT5a structure derives from the protein without ab ound ligand. In contrast, in the crystal structure of tyrosine-phosphorylated STAT3h omodimers, [13] the important amino acid side chains adopt asuitable conformation for phos- Figure 3. A) Activity of 20 against wild-type and mutant STAT5p roteins. n.a.: not applicable. B) Dockingp ose of 20 into the X-ray structure of murine STAT5a (PBD 1Y1U). [23] SH2 domain shown in yellow,l inker domain in blue; the side chains of amino acids defineda sflexible in the dockinga re shown with carbon atomsi ng reen.Divergentamino acids in the SH2 domain defined as rigid are shownw ith carbon atomsi nyellow.Asn664 is replaced by Ser in humanS TAT5a.The Figure was generated using PyMOL. [17] phate binding ( Figure S4). The virtual screening of STAT3d escribed in this study wasa chieved using AutoDock Vina, [14] which in our experience works best with rigid proteins.I nc ontrast, STAT5a docking requires an approachw ith flexible amino acid side chains, such as AutoDock FR, [24] which was used for rationalizing the binding mode of 20 ( Figure 3B). However,A u-toDockF Ri sl ess adaptable to virtual screening of large chemical libraries than AutoDock Vina.
Phosphonates are typically not cell-permeable, being negatively charged at physiological pH. To mask its negative charges, 26 was converted to the pivaloyloxymethylp rodrug 27 ( Figure 5A). Pivaloyloxymethyl prodrugs of organic phosphates [4a-c] and phosphonates [27] are cleaved by intracellulare sterases,r eleasing the bioactive parent compounds. 27 was tested for its ability to prevent phosphorylation of STAT5a in K562 cells, ah uman chronic myelogenous leukemiac ell line. In these cells, both STAT5a and STAT5b are constitutively phosphorylated on Tyr694 (STAT5a) and Tyr699 (STAT5b), respectively,b yB cr-Abl. Since phosphorylation is dependento nt he function of the SH2 domain, aS TAT5S H2 domain inhibitor will reduce the ability of STAT5t ob et yrosine phosphorylatedb y Bcr-Abl( Figure 5B). Commercially availablea ntibodies that recognize STAT5a/b only in their tyrosine phosphorylated state do not differentiate between the two STAT5p roteins, so we ectopically expressed fusion proteins of either STAT5a or STAT5b and GFP to distinguish unambiguously between the two STAT5 proteins. [4a-c] In the presence of the prodrug 27,p hosphorylation of STAT5a Tyr694 in the STAT5a-GFP fusion protein wasi nhibitedi nadose-dependent manner( Figure 5C,E). In contrast, phosphorylation of Tyr699 in the STAT5b-GFP construct was only minimally affected ( Figure 5D,F), showing that the selectivity of 26 for STAT5a over STAT5b is maintained in the cellular environment. In contrast, the STAT5i nhibitor AC-4-130, [28] for which no preference for either STAT5p rotein has been reported, and the Bcr-Abl inhibitor imatinib [29] do not discriminate between STAT5a and STAT5b in this assay ( Figure S8). Phosphorylation of endogenousS TAT5a/b was reduced to as ubstantially lower degree than that of STAT5a-GFP( Figure S9) by 27.T his is consistent with our previous observations, which indicate that the majority of endogenous phospho-STAT5i n K562 cells is STAT5b. [4a-c] In conclusion, we present docking-baseds creening of in silico O-phosphorylated natural product-derived fragments as an ovel method for identifying lead structures for the development of inhibitors of phosphorylation-dependentp rotein-protein interaction domains, which are of crucial importance to cellular signaling.W hereas virtuals creening of phosphonates and phosphates against ap hosphorylation-dependent protein-protein interaction domain has been reported, [30] our work represents the first case in which the virtual screening library itself is generatedb yi ns ilico O-phosphorylation. Application of this concept to the STAT3S H2 domain resulted in the moderate STAT3i nhibitor 1,w hich was then discovered to preferentially target STAT5. Analysis of structure-activity relationships led to the developmento f20,t he first inhibitor of STAT5a which displays high selectivityo ver STAT5b and other STAT family members. We dubbed 20 "Stafia-1" (STAT five ai nhibitor 1). The use of wild-type andp oint mutant STAT5p roteins demonstrated that both the SH2 domain and Trp566 in the adjacent linker domain contribute to selectiver ecognition of Stafia-1 by STAT5a. The cell-permeable prodrug 27,b ased on the Stafia-1-derived monofluoromethylene phosphonate 26, inhibitedt yrosine phosphorylation of STAT5a with selectivity over STAT5b in culturedh uman leukemia cells, and represents av aluablet ool to define the non-redundant molecular functions of the two highly homologous transcription factorsi n tumor cells. [7] Selective inhibition of STAT5a by 27,e specially in direct comparison with selectivei nhibitiono fS TAT5b by catechol bisphosphate-based prodrugs such as Pomstafib-2, [4c] would allow for dissection of the target genes of STAT5a and STAT5b with high temporal control. [7] Our data provide the first demonstration that selective targeting of STAT5a overS TAT5b is feasibleb oth in vitro and in cells.