Selective Inhibition of STAT3 with Respect to STAT1: Insights from

Aug 1, 2016 - Analysis of the docking results was performed with custom software based on the Python binding of the Pteros molecular modeling library...
0 downloads 12 Views 9MB Size
Subscriber access provided by SUNY DOWNSTATE

Article

Selective Inhibition of STAT3 with Respect to STAT1: Insights from Molecular Dynamics and Ensemble Docking Simulations Semen O. Yesylevskyy, Christophe Ramseyer, Marc Pudlo, Jean-Rene Pallandre, and Christophe Borg J. Chem. Inf. Model., Just Accepted Manuscript • DOI: 10.1021/acs.jcim.6b00198 • Publication Date (Web): 01 Aug 2016 Downloaded from http://pubs.acs.org on August 2, 2016

Just Accepted “Just Accepted” manuscripts have been peer-reviewed and accepted for publication. They are posted online prior to technical editing, formatting for publication and author proofing. The American Chemical Society provides “Just Accepted” as a free service to the research community to expedite the dissemination of scientific material as soon as possible after acceptance. “Just Accepted” manuscripts appear in full in PDF format accompanied by an HTML abstract. “Just Accepted” manuscripts have been fully peer reviewed, but should not be considered the official version of record. They are accessible to all readers and citable by the Digital Object Identifier (DOI®). “Just Accepted” is an optional service offered to authors. Therefore, the “Just Accepted” Web site may not include all articles that will be published in the journal. After a manuscript is technically edited and formatted, it will be removed from the “Just Accepted” Web site and published as an ASAP article. Note that technical editing may introduce minor changes to the manuscript text and/or graphics which could affect content, and all legal disclaimers and ethical guidelines that apply to the journal pertain. ACS cannot be held responsible for errors or consequences arising from the use of information contained in these “Just Accepted” manuscripts.

Journal of Chemical Information and Modeling is published by the American Chemical Society. 1155 Sixteenth Street N.W., Washington, DC 20036 Published by American Chemical Society. Copyright © American Chemical Society. However, no copyright claim is made to original U.S. Government works, or works produced by employees of any Commonwealth realm Crown government in the course of their duties.

Page 1 of 36

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Journal of Chemical Information and Modeling

Selective Inhibition of STAT3 with Respect to STAT1: Insights from Molecular Dynamics and Ensemble Docking Simulations Semen O. Yesylevskyy 1*, Christophe Ramseyer 2, Marc Pudlo3, Jean-René Pallandre 4 and Christophe Borg 4 1

Institute of Physics, National Academy of Sciences of Ukraine, Prospect Nauki, 46, Kyiv, 03039, Ukraine.

2

Laboratoire Chrono Environnement UMR CNRS 6249, Faculté des Sciences et Techniques, La Bouloie, Université Bourgogne Franche-Comté, 25030, Besançon Cedex, France. 3

Fonctions et Dysfonctions Epitheliales - EA 4267, Universite de Bourgogne Franche-Comté,

UFR Sciences Medicales et Pharmaceutiques, 19 rue Ambroise Pare, 25030 BESANCON cedex, France. 4

Inserm UMR 1098, EFS Bourgogne Franche Comté, Université Bourgogne Franche-Comté, IFR133, 8 rue du Dr Girod, 25020 Besançon, France.

Keywords STAT3, STAT1, selective inhibitors, molecular dynamics, ensemble docking, selectivity by distraction.

ACS Paragon Plus Environment

1

Journal of Chemical Information and Modeling

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 2 of 36

Abstract

STAT3 protein, which is known to be involved in cancer development, is a promising target for anti-cancer therapy. Successful inhibitors of STAT3 should not affect an activity of closely related protein STAT1, which makes their development challenging. The mechanisms of selectivity of several existing STAT3 inhibitors are not clear. In this work we studied molecular mechanisms of selectivity of thirteen experimentally tested STAT3 inhibitors by means of extensive molecular dynamics and ensemble docking simulations. It is shown that all studied inhibitors bind to the large part of protein surface in unspecific statistical manner. The binding to dimerization interface of the SH2 domain, which is usually considered as the main target region, is not energetically preferable. Binding in this region is remarkably similar for STAT1 and STAT3 proteins and can not explain experimentally observed selectivity towards STAT3. We propose new mechanism of selectivity called “selectivity by distraction” for existing STAT3 inhibitors. This mechanism is based on equilibrium statistical partitioning of inhibitor molecules between protein domains. The unspecific binding of inhibitors to the DNA-binding and the coilcoil domains is stronger in STAT1 in comparison to STAT3 while the energies of their binding to SH2 domains are comparable. This “distracts” inhibitor molecules from the SH2 domain of STAT1 and leads to higher effective concentration of inhibitors in the vicinity of SH2 domain of STAT3.

Introduction Human signal transducer and activator of transcription 3 protein (STAT3) is known to be involved in cancer development by promoting cancer cell growth, survival, angiogenesis, and immune evasion 1, 2. The active form of STAT3 is a dimer, which forms after phosphorylation of

ACS Paragon Plus Environment

2

Page 3 of 36

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Journal of Chemical Information and Modeling

the C-terminal phosphotyrosine (pTyr) loops of the monomers 3. The dimerization interface is located on the SH2 domain of STAT3, which binds to the pTyr loop of the other monomer in symmetric reciprocal manner. SH2 domain is identified in numerous proteins with diverse functions 4. This domain selectively binds phosphotyrosine peptides in many signaling proteins to mediate cellular pathways that are implicated in cancer, inflammation, allergy, and many other diseases 5. Suppression of STAT3 activity inhibits growth and induces apoptosis in vitro in cancer cell cultures and in vivo in animal models 2. In contrast, homologous and structurally similar STAT1 protein is thought to play an opposite role in the cancerogenesis 6. STAT1 may also be neutral in terms of cancer development but plays significant role in the cell growth, differentiation, inflammation and immune response 1. That is why any prospective anti-cancer drugs, which inhibit STAT3, should not suppress the activity of STAT1. The most promising way of inhibiting STAT3 protein to date is suppressing its dimerization by targeting the SH2 domain. However, close sequence and structure similarity of the SH2 domains of STAT1 and STAT3 make selective inhibition of STAT3 a challenging task. The first proposed inhibitors of STAT3 were peptides and phosphopeptides, which mimic the amino acid sequence of STAT3 pTyr loop interacting with the SH2 domain. However, such peptides have low stability and cell permeability, which stimulated development of non-peptidic inhibitors. There is a number of such inhibitors of very different chemical nature proposed to date 7-9. Recently new series of selective STAT3 inhibitors containing aminotetrazole group were proposed by the authors 10. Although many selective STAT3 inhibitors exist the molecular mechanism of their selectivity is still under discussion. The selectivity which is observed is puzzling since the sequences of

ACS Paragon Plus Environment

3

Journal of Chemical Information and Modeling

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 4 of 36

SH2 domains of STAT3 and STAT1 proteins are very similar. Particularly, the sequences of the SH2 binding spots, which are involved in the dimerization process, are 100% identical

11

.

Existing homology modeling studies reveal minor differences in structure of the dimerization interfaces of STAT1 and STAT3

11

, however, their significance is hard to evaluate for static

protein structures. Despite large number of known STAT3 inhibitors there is no dedicated in silico study, which address the molecular mechanism of selectivity for known STAT3 inhibitors on the atomistic level. In this work, we address this problem by studying the binding of thirteen selective nonpeptidic inhibitors of STAT3 to STAT3 and STAT1 proteins (Fig. 1) by means of comprehensive ensemble docking simulations. Eight well-known STAT3 inhibitors sta-21 12, bp1-102 13, S3I-1757 14, S3I-201 15, S3I-M2001 16, FLLL32 17, azepine derivative 1 from Leung et. al

18

(abbreviated as azepine-1) and benzofuran derivative 1 from Liu et. al

19

(abbreviated as

benzofuran-1) are used as well as the family of five inhibitors containing aminotetrazole group recently designed by the authors (compounds 23, 35, 5, 42, 43 from 10). The later compounds are shown to selectively inhibit STAT3 according to in-vitro luciferase reporter assay tests (see Table S1).

ACS Paragon Plus Environment

4

Page 5 of 36

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Journal of Chemical Information and Modeling

Figure 1. Structures of used inhibitors.

The flexibility and dynamics of STAT1 and STAT3 monomers in water environment are taken into account by means of all-atom molecular dynamics (MD) simulations. The binding of inhibitors to all protein domains namely SH2 domain, DNA binding site and the coil-coil domain is analyzed (Fig. 2). Although all studied inhibitors are believed to bind preferably to dimerization interface of the SH2 domain of STAT3 we show that their binding is surprisingly unspecific. Particularly, there is no pronounced difference in the binding to dimerization

ACS Paragon Plus Environment

5

Journal of Chemical Information and Modeling

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 6 of 36

interfaces of STAT1 and STAT3 in all studied compounds. However, the selectivity to STAT3 could be caused by unusual statistical mechanism of partitioning inhibitor moieties between different protein domains called “selectivity by distraction”.

SH2 domain

Linker domain

DNA-binding domain

Coil-coil domain

Figure 2. The scheme of the domain structure of STAT1 and STAT3 proteins. The crystal structure of human STAT1 monomer (PDB code 1BF5) is used for illustrative purposes.

ACS Paragon Plus Environment

6

Page 7 of 36

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Journal of Chemical Information and Modeling

Methods The human STAT1 protein is crystallized as a dimer bound to DNA (PDB code 1BF5). This structure is not complete and misses several residues located in the flexible loops including the phosphotyrosine loop. There is no crystal structure of human STAT3 available to date, thus homology modeling should be used to construct the model of human STAT3. Two possible templates for such homology modeling are currently available – mouse STAT3 dimer bound to DNA (1BG1) or human STAT1 dimer (1BF5). Since there is no experimental evidence about structural similarity of native human STAT3 with these candidate proteins we decided to use both of them and to produce two models based on either human STAT1 template (referred as STAT3h) or mouse STAT3 template (referred as STAT3m). The homology modeling was performed with the SWISS-MODEL server

20

. The same

server was used to reconstruct the gaps in the incomplete human STAT1 structure 1BF5 based on its full-length sequence. All three protein models (STAT1, STAT3h and STAT3m) were subjected to molecular dynamics simulations in water in the NPT ensemble at the temperature of 300 K and the pressure of 1 atm. GROMACS 4.6.5 software suite 21 with GROMOS 54A7force field 22 was used for all simulations. The proteins were energy minimized and subjected to 20 ns MD simulations with constrained backbone followed by unconstrained 60 ns MD runs. The equilibration was monitored by the RMSD of all protein atoms after alignment of trajectories by their Cα atoms. For all proteins the equilibrium was reached after 30-40 ns of simulations (Fig. 3). Last 10 ns of each trajectory were used to extract 200 frames at 5 ps intervals. These frames were used as representative structures for the ensemble docking simulations.

ACS Paragon Plus Environment

7

Journal of Chemical Information and Modeling

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 8 of 36

The docking procedure was performed as following. The structures of all inhibitors were optimized in Gaussian09 at the B3LYP/6-31++G(d) level of theory. The ligands and the corresponding protein structures were prepared for docking using the MGLTools-1.5.6. The docking was performed with the Autodock Vina

23

software with the default scoring function.

The center and the size of the search volume were adjusted with custom scripts to match different parts of the protein. The exhaustiveness of the search was adjusted to be proportional to the search volume in each case. Analysis of the docking results was performed with custom software based on the Python binding of the Pteros molecular modelling library

24, 25

. For each

docking simulations, 9 top ranked poses were recorded. The following parts of the proteins were used for docking independently: the SH2 domain (search centered at the center of masses of residues ASN 574, GLU 587, ARG 602, GLU 605, TRP 616, PHE 628, VAL 631, TYR 651 for STAT1; LYS 591, GLU 594, ARG 609, GLU 612, TRP 623, ILE 634, VAL 637, TYR 657 for STAT3, search volume 50x50x50 nm for the whole domain or 20x20x20 nm for dimerization interface only), the DNA-binding domain (search centered at the center of masses of residues in contact with DNA in the crystal structure, search volume 50x50x50 nm) and the coil-coil domain (search centered at the residue 155 for STAT1 and 157 for STAT3, search volume 50x50x50 nm). Due to inadequate sampling of possible conformations of the phosphotyrosine loop (see below) it was excluded from docking simulations. As a result, for each protein domains and for each compound 1800 docking poses were obtained (200 MD frames, 9 poses per frame).

Results

ACS Paragon Plus Environment

8

Page 9 of 36

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Journal of Chemical Information and Modeling

Protein dynamics The results of MD simulations show that the structures of STAT1 and STAT3 proteins are rather rigid. There are no significant interdomain motions and no visible changes in the inner parts of individual domains. The whole protein RMSD stabilizes after ~30 ns in all simulations (Fig. 3) at the level of 0.5-0.65 nm. This is rather small value for such large protein (>700 residues), which indicates remarkable stability of the overall protein structure. The amplitude of RMSD fluctuations is smallest in STAT1 and somewhat larger in STAT3m and STAT3h. Both models of STAT3 show similar overall dynamics. The difference in fluctuations between STAT3m and STAT3h homology models is comparable with random differences between independent MD runs of the same system thus both models could be considered sufficiently stable.

Figure 3. Evolution of Cα root mean square deviations of the studied proteins in course of MD simulations.

ACS Paragon Plus Environment

9

Journal of Chemical Information and Modeling

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

A

Page 10 of 36

B

C

Figure 4. Initial (blue) and final (red) conformations of the phosphotyrosine loop in STAT1 (A), STAT3m (B) and STAT3h (C). Phosphotyrosine residue is shown in space-filling representation. Orientations and scales are different in different panels.

The phosphotyrosine loop of one monomer is bound to the second monomer in dimeric crystal structure of STAT1 and STAT3. In our MD simulations only one monomer is present, which means that the crystal conformation of the phosphotyrosine loop is no longer stable. Indeed, this loop is very dynamic and adopts different structures in different MD simulations (Fig. 4). Particularly, the phosphotyrosine loop of STAT1 losses helical structure in few nanoseconds and binds to the surface of the SH2 domains as disordered coil up to the end of simulation. In

ACS Paragon Plus Environment

10

Page 11 of 36

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Journal of Chemical Information and Modeling

STAT3m the loops is disordered in the initial structure but in 20-30 ns it adopts the conformation of the anti-parallel β-hairpin, which remains stable for the rest of simulation. In STAT3h the loops is helical in the beginning (it is based on the same template as STAT1), but in contrast to STAT1, it does not unfold and maintains helical structure for the whole simulation. However, its orientation changes significantly and it partially binds to the surface of the SH2 domain after 1020 ns. It is evident that exhaustive sampling of possible conformations of this very flexible loop is not possible in atomistic MD simulations in reasonable time. It is also clear that crystal structure of the dimer does not reflect conformation of the phosphotyrosine loop in the monomeric protein. Thus the phosphotyrosine loop was excluded from subsequent docking simulations to ensure that docking results are not biased by its instantaneous conformation.

Ensemble docking Due to the large size of the proteins investigated in the present work, blind docking to the whole protein is ineffective and time consuming. Thus, the ligands were docked separately to different structural domains of the proteins. The results show that the binding of all inhibitors to STAT1, STAT3m and STAT3h is surprisingly unspecific. There are no specific well-defined binding sites, which have pronounced selective affinity to any of the inhibitors. In contrast, the inhibitors bind rather randomly to a wide variety of different sites on the protein surface with comparable binding scores. The best binding positions and scores for the same inhibitor change without visible trend from one MD frame to the other. Fig. 5 shows the structure of STAT1 with superimposed best docking positions of all inhibitors for all performed docking simulations. It is clearly seen that there are many large fuzzy areas of unspecific binding, which cover the surfaces of all protein domains. None of them could be classified as a true binding site, however. The best

ACS Paragon Plus Environment

11

Journal of Chemical Information and Modeling

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 12 of 36

binding scores for all such areas are between -7 and -12 kcal/mol depending on the inhibitor and the protein domain (Fig. 6).

Figure 5. Average structure of STAT1 protein with superimposed best docking positions of all inhibitors in all performed docking simulations. Docking positions are shown after the least square fit of particular frame to the starting structure. Due to very small overall flexibility of the protein such representation provides reliable picture of binding locations. The SH2 domain is yellow, the linking domain is green, the DNA-binding domain is red, the coil-coil domain is purple. The inhibitors are shown in wireframe representation.

Proof of unspecific binding

ACS Paragon Plus Environment

12

Page 13 of 36

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Journal of Chemical Information and Modeling

Although the dimerization interface of the SH2 domain is usually considered as a target for majority of known selective STAT3 inhibitors, we did not find any well-defined preferential binding in this area. In contrast, for all studied inhibitors the absolute best binding scores are observed in the DNA-binding domain (see Fig. 6), while the best scores on the dimerization interface are usually 1-2 kcal/mol worse. Due to statistical nature of ensemble docking such result could be caused by accidental appearance of single unusual protein conformation in the sampled trajectory frames, which corresponds to the best binding score but appears with extremely small probability in reality. In order to exclude such possibility we analyzed average binding scores for each inhibitor computed over nine top-ranked docking poses from each trajectory frame. Obtained trends are almost identical to those in Fig. 6 (see Fig. S1) which means that the absence of specific binding of inhibitors to SH2 domain is not a sampling artifact.

Figure 6. Best binding scores (kcal/mol) in different protein domain for all studied inhibitors. A detailed comparison of individual binding “spots” distributed over the protein surface in different proteins appears to be challenging due to the sampling of different sets of protein

ACS Paragon Plus Environment

13

Journal of Chemical Information and Modeling

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 14 of 36

structures in individual MD simulations. However it is possible to compare the binding of different inhibitors to the same protein. In order to do this we computed probabilities of binding of each inhibitor to each of protein residues over used set of trajectory frames. Obtained binding probability profiles are rather hard to analyze visually (see figures S2-S4) thus we computed pair correlations between them and performed a cluster analysis to reveal similarities quantitatively. Obtained results are shown as the dendrograms in Fig. 7, which visualize similarities between the binding probability profiles in the form of hierarchical tree. The details of clustering on such trees, which are different for different proteins, are not important for us. In contrast, the position of the first branching is of interest because it shows correlation between least similar binding probability profiles.

Figure 7. Clustering analysis of binding probability profiles of different inhibitors (shown in figures S2-S4). Agglomerative clustering with average linkage is used. “Distance” between the clusters is an average pair correlation coefficient between the binding probability profiles of inhibitors included into the clusters.

ACS Paragon Plus Environment

14

Page 15 of 36

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Journal of Chemical Information and Modeling

Figure 7 shows that in the case of STAT3h the binding probability profiles of all studied inhibitors are very similar (the first branching occurs at ~0.85). In the case of STAT3m the first branching occurs at ~0.68, however there two distinct clusters of inhibitors with very high internal correlations of ~0.87-0.88. The first cluster contains compounds 42, sta-21 and benzofuran-1, while the second contains all other compounds. In the case of STAT1 all inhibitors except azepine-1 have extremely similar binding profiles with minimal correlation of ~0.88. Azepine-1 is very different from other inhibitors in this case (correlation with other compounds is only ~0.55) which is likely to be an artifact because such behavior is not observed in STAT3m and STAT4h. It is possible to conclude that for each of studied proteins all inhibitors bind to the same set of residues with almost the same probability regardless of their chemical structure. The correlations between their binding profiles are higher than 0.85 with only few exceptions, which could be attributed to sampling and analysis artifacts. This means that the binding of all studied inhibitors is remarkably unspecific with no visible correlation with their chemical structure.

Restricted docking to dimerization interface of SH2 domain

ACS Paragon Plus Environment

15

Journal of Chemical Information and Modeling

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 16 of 36

Figure 8. Probabilities of binding to particular amino acids on the dimerization interface of the SH2 domain in restricted docking simulations. If the docking volume is restricted to the dimerization interface of the SH2 domain the binding of all inhibitors still remains remarkably unspecific in terms of their exact positions and orientations. Probabilities of binding to particular residues on dimerization interface of SH2 domain are consistent for all compounds within the same protein (Fig. 8). It is clearly seen that for each particular protein the differences in binding probabilities between different inhibitors are barely visible despite their strikingly different chemical structures. All inhibitors bind to the same set of residues on the dimerization interface with only minor quantitative variations. Some details of the probability profiles are different for STAT1, STAT3m and STAT3h but these differences are observed consistently for all inhibitors. Thus, these differences could be attributed to the sampling variations between MD trajectories. Pronounced similarity of binding probabilities for all studied inhibitors clearly indicates unspecific nature of their binding, which lack significant correlation with the chemical structure of particular compound.

ACS Paragon Plus Environment

16

Page 17 of 36

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Journal of Chemical Information and Modeling

Restricted docking to the DNA-binding domain The major point of interest in restricted docking to the DNA-binding domain is possible binding of inhibitors to the residues, which are involved in DNA recognition. For STAT1 these residues are MET 325, THR 327, HIS 328, LYS 336 - GLN340, ARG 378, GLU 411, LYS 413, ASN 414, ARG 418 – GLU 421, ILE 425 – THR 427, ILE 458 – GLN 463, GLU 563, LYS 567 (determined as the residues, which are within 5Å of any DNA atom in the crystal structure). For STAT3 corresponding residues are MET 329, MET 331, HIS 332, LYS 340 – GLN 344, ARG 382, GLU 415, ARG 417, ASN 425 – ASP 427, ILE 431 – GLU 434, ILE 464 – GLN 469, LYS 574. Analysis of the binding probabilities shows that all studied inhibitors never bind to these residues (binding probability