Towards the Inhibition of Protein–Protein Interactions (PPIs) in STAT3: Insights into a New Class of Benzothiadiazole Derivatives

Signal transducer and activator of transcription 3 (STAT3) is a validated anticancer target due to the relationship between its constitutive activation and malignant tumors. Through a virtual screening approach on the STAT3-SH2 domain, 5,6-dimethyl-1H,3H-2,1,3-benzothiadiazole-2,2-dioxide (1) was identified as a potential STAT3 inhibitor. Some benzothiadiazole derivatives were synthesized by employing a versatile methodology, and they were tested by an AlphaScreen-based assay. Among them, benzosulfamide 1 showed a significant activity with an IC50 = 15.8 ± 0.6 µM as a direct STAT3 inhibitor. Notably, we discovered that compound 1 was also able to interact with cysteine residues located around the SH2 domain. By applying mass spectrometry, liquid chromatography, NMR, and UV spectroscopy, an in-depth investigation was carried out, shedding light on its intriguing and unexpected mechanism of interaction.


Introduction
Since the discovery of the relationship between the constitutive activation of signal transducer and activator of transcription 3 (STAT3) and malignant tumors, the validation of STAT3 as a target has been supported by a large number of studies. Various compounds with STAT3 inhibitory activity have been reported in recent years, including potent inhibitors such as Stattic, S3I-201, and erasin [1][2][3][4]. Nevertheless, these molecules are mostly at the experimental stage, and only few are in clinical trials [5]. Hence, the challenge is the discovery of new selective candidates with high potency and in vivo activity.
Since a direct inhibitory approach should be preferred, we addressed our efforts to the disruption of STAT3 homodimerization, an essential step in its activation, targeting the protein-protein interactions (PPIs). The large surface area of STAT3 and the chemistry of these interactions [6], which are very different from those of better-known targets such as enzymes and G-protein-coupled receptors, could represent significant hurdles, although a number of successful examples have demonstrated that it is a hot spot should compete with the original protein partner of the PPI, thus resulting in disruption of its function, the aim of our ongoing studies [8][9][10][11][12][13] is the identification of small molecules that are able to bind the Src homology 2 (SH2) domain of STAT3, preventing STAT3 dimerization and thus its activation. With this purpose, in the present work, some compound libraries were analyzed by means of a structure-based virtual screening performed on the STAT3-SH2 domain (see the Molecular Modeling section). Among the candidates that reached the best scores in the docking study, we decided to focus our attention on the compounds that bear a sulfone moiety, which is a common feature in some STAT3-SH2 inhibitors reported in the literature [1,2]. In order to support the docking results, the selected compounds were either purchased or synthesized, and their ability to interact with the STAT3-SH2 domain was evaluated via an in vitro binding test (AlphaScreen-based assay) at 30 µM; the most interesting results are reported in Table 1. Hence, we identified the benzothiadiazole derivative 1 as a promising hit compound that carries a chemically accessible moiety for further development.
For the synthesis of 1a-i, we employed an adaptable methodology [14] that also considered a diversity-oriented approach using suitable monoprotected intermediates. For derivatives 2a,b, 3, and 4, our synthetic strategy was planned to allow for the introduction of different substituents on compound (1) by the direct alkylation of the sulfamide moiety.
Herein, we describe the in silico-based rational design, synthesis, characterization, and interaction with the STAT3-SH2 domain of new inhibitors bearing the benzosulfamide scaffold. Furthermore, since several STAT3 inhibitors have been reported as irreversible cysteine binders (e.g., Stattic and S3I-201) [15][16][17][18], we decided to evaluate the activity of compound 1 versus the cysteine residues located around the STAT3-SH2 domain (Cys468, Cys542, Cys550, Cys687, and Cys712) and to shed light on its mechanism of interaction by MS, UV, LC, and NMR studies. a hot spot should compete with the original protein partner of the PPI, thus resulting in disruption of its function, the aim of our ongoing studies [8][9][10][11][12][13] is the identification of small molecules that are able to bind the Src homology 2 (SH2) domain of STAT3, preventing STAT3 dimerization and thus its activation. With this purpose, in the present work, some compound libraries were analyzed by means of a structure-based virtual screening performed on the STAT3-SH2 domain (see the Molecular Modeling section). Among the candidates that reached the best scores in the docking study, we decided to focus our attention on the compounds that bear a sulfone moiety, which is a common feature in some STAT3-SH2 inhibitors reported in the literature [1,2]. In order to support the docking results, the selected compounds were either purchased or synthesized, and their ability to interact with the STAT3-SH2 domain was evaluated via an in vitro binding test (AlphaScreen-based assay) at 30 µM; the most interesting results are reported in Table 1. Hence, we identified the benzothiadiazole derivative 1 as a promising hit compound that carries a chemically accessible moiety for further development.
For the synthesis of 1a-i, we employed an adaptable methodology [14] that also considered a diversity-oriented approach using suitable monoprotected intermediates. For derivatives 2a,b, 3, and 4, our synthetic strategy was planned to allow for the introduction of different substituents on compound (1) by the direct alkylation of the sulfamide moiety.
Herein, we describe the in silico-based rational design, synthesis, characterization, and interaction with the STAT3-SH2 domain of new inhibitors bearing the benzosulfamide scaffold. Furthermore, since several STAT3 inhibitors have been reported as irreversible cysteine binders (e.g., Stattic and S3I-201) [15][16][17][18], we decided to evaluate the activity of compound 1 versus the cysteine residues located around the STAT3-SH2 domain (Cys468, Cys542, Cys550, Cys687, and Cys712) and to shed light on its mechanism of interaction by MS, UV, LC, and NMR studies. a hot spot should compete with the original protein partner of the PPI, thus resulting in disruption of its function, the aim of our ongoing studies [8][9][10][11][12][13] is the identification of small molecules that are able to bind the Src homology 2 (SH2) domain of STAT3, preventing STAT3 dimerization and thus its activation. With this purpose, in the present work, some compound libraries were analyzed by means of a structure-based virtual screening performed on the STAT3-SH2 domain (see the Molecular Modeling section). Among the candidates that reached the best scores in the docking study, we decided to focus our attention on the compounds that bear a sulfone moiety, which is a common feature in some STAT3-SH2 inhibitors reported in the literature [1,2]. In order to support the docking results, the selected compounds were either purchased or synthesized, and their ability to interact with the STAT3-SH2 domain was evaluated via an in vitro binding test (AlphaScreen-based assay) at 30 µM; the most interesting results are reported in Table 1. Hence, we identified the benzothiadiazole derivative 1 as a promising hit compound that carries a chemically accessible moiety for further development.
For the synthesis of 1a-i, we employed an adaptable methodology [14] that also considered a diversity-oriented approach using suitable monoprotected intermediates. For derivatives 2a,b, 3, and 4, our synthetic strategy was planned to allow for the introduction of different substituents on compound (1) by the direct alkylation of the sulfamide moiety.
Herein, we describe the in silico-based rational design, synthesis, characterization, and interaction with the STAT3-SH2 domain of new inhibitors bearing the benzosulfamide scaffold. Furthermore, since several STAT3 inhibitors have been reported as irreversible cysteine binders (e.g., Stattic and S3I-201) [15][16][17][18], we decided to evaluate the activity of compound 1 versus the cysteine residues located around the STAT3-SH2 domain (Cys468, Cys542, Cys550, Cys687, and Cys712) and to shed light on its mechanism of interaction by MS, UV, LC, and NMR studies.

III
Molecules 2020, 25, x FOR PEER REVIEW 2 of 29 a hot spot should compete with the original protein partner of the PPI, thus resulting in disruption of its function, the aim of our ongoing studies [8][9][10][11][12][13] is the identification of small molecules that are able to bind the Src homology 2 (SH2) domain of STAT3, preventing STAT3 dimerization and thus its activation. With this purpose, in the present work, some compound libraries were analyzed by means of a structure-based virtual screening performed on the STAT3-SH2 domain (see the Molecular Modeling section). Among the candidates that reached the best scores in the docking study, we decided to focus our attention on the compounds that bear a sulfone moiety, which is a common feature in some STAT3-SH2 inhibitors reported in the literature [1,2]. In order to support the docking results, the selected compounds were either purchased or synthesized, and their ability to interact with the STAT3-SH2 domain was evaluated via an in vitro binding test (AlphaScreen-based assay) at 30 µM; the most interesting results are reported in Table 1. Hence, we identified the benzothiadiazole derivative 1 as a promising hit compound that carries a chemically accessible moiety for further development.
For the synthesis of 1a-i, we employed an adaptable methodology [14] that also considered a diversity-oriented approach using suitable monoprotected intermediates. For derivatives 2a,b, 3, and 4, our synthetic strategy was planned to allow for the introduction of different substituents on compound (1) by the direct alkylation of the sulfamide moiety.
Herein, we describe the in silico-based rational design, synthesis, characterization, and interaction with the STAT3-SH2 domain of new inhibitors bearing the benzosulfamide scaffold. Furthermore, since several STAT3 inhibitors have been reported as irreversible cysteine binders (e.g., Stattic and S3I-201) [15][16][17][18], we decided to evaluate the activity of compound 1 versus the cysteine residues located around the STAT3-SH2 domain (Cys468, Cys542, Cys550, Cys687, and Cys712) and to shed light on its mechanism of interaction by MS, UV, LC, and NMR studies.
For the synthesis of 1a-i, we employed an adaptable methodology [14] that also considered a diversity-oriented approach using suitable monoprotected intermediates. For derivatives 2a,b, 3, and 4, our synthetic strategy was planned to allow for the introduction of different substituents on compound (1) by the direct alkylation of the sulfamide moiety.

Molecular Modeling-Design and Chemical Space Exploration
In order to identify new direct STAT3 inhibitors, molecular modeling studies were carried out. In particular, a virtual screening approach was applied to compound databases with the aim to select a series of candidates endowed with the best binding scores. Therefore, we performed a structure-based virtual screening, choosing STAT3 in complex with a DNA fragment (Protein Data Bank (PDB) ID 1BG1) [19] as a model and employing a set of databases of small molecules, namely: AKos (544391 molecules), ChemPDB (4009), ChemBank (2344), Kyoto Encyclopedia of Genes and Genomes -Compound (KEGG, 10005), National Cancer Institute Anti-HIV( NCI's Anti-HIV, 42689), and National Cancer Institute (NCI, 15237).
Each molecule was ranked on the basis of the docking score (interaction energy of AutoDock 4) [20] of the best solution. In this way, a list of compounds was obtained in which the first and the last molecules were, respectively, the best and the worst interacting ones. This list made the next step of the analysis easier because only the first 500 compounds for each database were further considered. Hence, the best molecules were classified on the basis of the most recurrent moieties: (e.g., sulfonic acids, sulfonamides, sulfamates, phosphates, and diketones; Tables S1-S5, Supplementary Materials). This kind of clustering was performed by encoding the moieties into SMARTS patterns and performing queries on the databases of the top-ranked poses, as implemented in the VEGA ZZ program. These results were very interesting because the functional groups included negatively charged atoms, such as oxygens, which can mimic the phosphate moiety of Tyr705 (pTyr-705), upon interaction with positively charged amino acids. A further selection was performed based on the commercial availability, synthetic accessibility, and size of the molecules. In particular, small molecules were privileged because, in the future, they may be included in peptidic derivatives to improve their selectivity. Therefore, drawing inspiration from compounds reported in the literature [1,2], we decided to focus on the cyclic sulfonamide moiety, ultimately selecting compounds with a molecular weight lower than 300 Da. The biological activity of several candidates belonging to the NCI collection was evaluated by an AlphaScreen-based assay, and among the most interesting hits, compound 1 emerged as a promising lead for further development (Table 1). The analysis of its docking pose (see Figure S1) revealed that compound 1 may interact with the SH2 domain through the insertion of its sulfonamide moiety into the polar pocket lined by Arg-609 and Lys-591; moreover, it showed that the compound occupies most of the pocket volume normally filled by pTyr-705 in the dimerized STAT3. Starting from this

Molecular Modeling-Design and Chemical Space Exploration
In order to identify new direct STAT3 inhibitors, molecular modeling studies were carried out. In particular, a virtual screening approach was applied to compound databases with the aim to select a series of candidates endowed with the best binding scores. Therefore, we performed a structure-based virtual screening, choosing STAT3 in complex with a DNA fragment (Protein Data Bank (PDB) ID 1BG1) [19] as a model and employing a set of databases of small molecules, namely: AKos (544391 molecules), ChemPDB (4009), ChemBank (2344), Kyoto Encyclopedia of Genes and Genomes-Compound (KEGG, 10005), National Cancer Institute Anti-HIV( NCI's Anti-HIV, 42689), and National Cancer Institute (NCI, 15237).
Each molecule was ranked on the basis of the docking score (interaction energy of AutoDock 4) [20] of the best solution. In this way, a list of compounds was obtained in which the first and the last molecules were, respectively, the best and the worst interacting ones. This list made the next step of the analysis easier because only the first 500 compounds for each database were further considered. Hence, the best molecules were classified on the basis of the most recurrent moieties: (e.g., sulfonic acids, sulfonamides, sulfamates, phosphates, and diketones; Tables S1-S5, Supplementary Materials). This kind of clustering was performed by encoding the moieties into SMARTS patterns and performing queries on the databases of the top-ranked poses, as implemented in the VEGA ZZ program. These results were very interesting because the functional groups included negatively charged atoms, such as oxygens, which can mimic the phosphate moiety of Tyr705 (pTyr-705), upon interaction with positively charged amino acids. A further selection was performed based on the commercial availability, synthetic accessibility, and size of the molecules. In particular, small molecules were privileged because, in the future, they may be included in peptidic derivatives to improve their selectivity. Therefore, drawing inspiration from compounds reported in the literature [1,2], we decided to focus on the cyclic sulfonamide moiety, ultimately selecting compounds with a molecular weight lower than 300 Da. The biological activity of several candidates belonging to the NCI collection was evaluated by an AlphaScreen-based assay, and among the most interesting hits, compound 1 emerged as a promising lead for further development (Table 1). The analysis of its docking pose (see Figure S1) revealed that compound 1 may interact with the SH2 domain through the insertion of its sulfonamide moiety into the polar pocket lined by Arg-609 and Lys-591; moreover, it showed that the compound occupies most of the pocket volume normally filled by pTyr-705 in the dimerized STAT3. Starting from this qualitative analysis of the binding mode, it seems reasonable to hypothesize that this molecule could act as a reversible inhibitor of STAT3 dimerization.

Chemistry
Compounds 1 and 1a-d were synthesized from the corresponding commercially available ortho-diamines through a direct cyclization in the presence of sulfamide and diglyme (Scheme 1) [21].
Molecules 2020, 25, x FOR PEER REVIEW 4 of 29 qualitative analysis of the binding mode, it seems reasonable to hypothesize that this molecule could act as a reversible inhibitor of STAT3 dimerization.
Compound 1 showed acidic properties (pKa = 6.88 ± 0.03) [22] and was fully characterized by X-ray diffraction. Its crystal structure is shown in Figure 2 as an ORTEP [23] drawing ( Figure 2). qualitative analysis of the binding mode, it seems reasonable to hypothesize that this molecule could act as a reversible inhibitor of STAT3 dimerization.
Product 1f was obtained starting from compound 1d, which was protected by treatment with PMB-Br and potassium carbonate to afford intermediate 13; its reduction with zinc and ammonium chloride gave the amine 14. The latter, through a reaction with acetic anhydride, afforded intermediate 15, which was deprotected with TFA to give product 1f (Scheme 3).
Intermediate 19, the starting reagent for the synthesis of products 1g-i, was obtained as reported in Scheme 4: the reaction of 2,5-dibromo-nitrobenzene with PMB-NH 2 led to intermediate 16, which was reduced to give the amine 17 in the presence of zinc and ammonium chloride. By treatment with sulfamide, 17 was cyclized, thus affording the benzothiadiazole derivative 18, which was subsequently protected with PMB-Br to yield compound 19 (Scheme 4). Product 1f was obtained starting from compound 1d, which was protected by treatment with PMB-Br and potassium carbonate to afford intermediate 13; its reduction with zinc and ammonium chloride gave the amine 14. The latter, through a reaction with acetic anhydride, afforded intermediate 15, which was deprotected with TFA to give product 1f (Scheme 3). Intermediate 19, the starting reagent for the synthesis of products 1g-i, was obtained as reported in Scheme 4: the reaction of 2,5-dibromo-nitrobenzene with PMB-NH2 led to intermediate 16, which was reduced to give the amine 17 in the presence of zinc and ammonium chloride. By treatment with sulfamide, 17 was cyclized, thus affording the benzothiadiazole derivative 18, which was subsequently protected with PMB-Br to yield compound 19 (Scheme 4). Product 1f was obtained starting from compound 1d, which was protected by treatment with PMB-Br and potassium carbonate to afford intermediate 13; its reduction with zinc and ammonium chloride gave the amine 14. The latter, through a reaction with acetic anhydride, afforded intermediate 15, which was deprotected with TFA to give product 1f (Scheme 3). Intermediate 19, the starting reagent for the synthesis of products 1g-i, was obtained as reported in Scheme 4: the reaction of 2,5-dibromo-nitrobenzene with PMB-NH2 led to intermediate 16, which was reduced to give the amine 17 in the presence of zinc and ammonium chloride. By treatment with sulfamide, 17 was cyclized, thus affording the benzothiadiazole derivative 18, which was subsequently protected with PMB-Br to yield compound 19 (Scheme 4).  Afterwards, intermediate 19 was reacted with ethyl acrylate in the presence of tri-o-tolylphosphine and tris(dibenzylideneacetone)dipalladium(0) to afford the ester 20. The latter was hydrolyzed under basic conditions to the corresponding acid 21, which was converted to the amides 22 and 23 using isopentylamine, HATU, and N,N-diisopropylethylamine or aniline, TBTU, and Nmethylmorpholine, respectively. By catalytic hydrogenation with 10% Pd/C, they were reduced and deprotected to give products 1g and 1h (Scheme 5). Products 2b and 3 were obtained from compounds 1b and 1a, respectively, through alkylation with iodomethane or benzyl bromide in the presence of potassium carbonate (Scheme 7).  Moreover, compound 19 was arylated through a Suzuki reaction with phenylboronic acid in the presence of tetrakis(triphenylphosphine)palladium(0) to give intermediate 24, which was then deprotected by treatment with TFA, thus affording product 1i (Scheme 6). and tris(dibenzylideneacetone)dipalladium(0) to afford the ester 20. The latter was hydrolyzed under basic conditions to the corresponding acid 21, which was converted to the amides 22 and 23 using isopentylamine, HATU, and N,N-diisopropylethylamine or aniline, TBTU, and Nmethylmorpholine, respectively. By catalytic hydrogenation with 10% Pd/C, they were reduced and deprotected to give products 1g and 1h (Scheme 5). Products 2b and 3 were obtained from compounds 1b and 1a, respectively, through alkylation with iodomethane or benzyl bromide in the presence of potassium carbonate (Scheme 7). Products 2b and 3 were obtained from compounds 1b and 1a, respectively, through alkylation with iodomethane or benzyl bromide in the presence of potassium carbonate (Scheme 7). and tris(dibenzylideneacetone)dipalladium(0) to afford the ester 20. The latter was hydrolyzed under basic conditions to the corresponding acid 21, which was converted to the amides 22 and 23 using isopentylamine, HATU, and N,N-diisopropylethylamine or aniline, TBTU, and Nmethylmorpholine, respectively. By catalytic hydrogenation with 10% Pd/C, they were reduced and deprotected to give products 1g and 1h (Scheme 5).  Finally, compound 6 was purchased from commercial sources.

Biological Evaluation
AlphaScreen is an in vitro competitive binding test used to identify compounds that are able to inhibit the binding of proteins containing an SH2 domain to their corresponding phosphopeptides (5-carboxyfluorescein (FITC)-GpYLPQTV for STAT3 and FITC-GpYDKPHVL for STAT1) ( Table 2). Compounds characterized by an interesting STAT3 affinity were further investigated for their selectivity versus STAT1, which shows a high degree of sequence similarity to STAT3 but an opposite physiological role [24,25], and Grb2, as a model for other SH2 domain-containing proteins.
The significant activity of compound 1, revealed at 30 µM (Table 2), prompted us to test this compound at different concentrations. Therefore, we calculated its IC50 values against STAT3 and STAT1 to be 15.8 ± 0.6 and ˃ 50 µM, respectively, while the inhibition versus Grb2 was 23% (at a 30 µM concentration).
To identify a trend in terms of structure-activity relationship, we firstly explored the role of the substituents on the benzene ring (1a-i). In this respect, we found that lipophilicity does not play an important role in modulating STAT3 inhibitory activity. Bulky lipophilic substituents at position 5 (derivatives 1g and 1h) induced a significant drop in the inhibition properties (Table 2). Among these derivatives, only 1d, bearing a NO2 group at position 5, exhibited a noticeable activity against STAT3. For this compound, we calculated the IC50 values against STAT3 and STAT1, which were 25.1 ± 2.6 µM and 15.9 ± 2.0 µM, respectively, revealing no selectivity for STAT3 inhibition. Then, we introduced different groups on the nitrogen atoms of the thiadiazole ring (2a,b, 3, and 4), in order to explore asymmetrical substitutions: The presence of the N-substituent did not improve activity. The enhanced bulkiness and hydrophobicity of these substituents compared to compound 1 did not lead to a better interaction.  Finally, compound 6 was purchased from commercial sources.

Biological Evaluation
AlphaScreen is an in vitro competitive binding test used to identify compounds that are able to inhibit the binding of proteins containing an SH2 domain to their corresponding phosphopeptides (5-carboxyfluorescein (FITC)-GpYLPQTV for STAT3 and FITC-GpYDKPHVL for STAT1) ( Table 2). Compounds characterized by an interesting STAT3 affinity were further investigated for their selectivity versus STAT1, which shows a high degree of sequence similarity to STAT3 but an opposite physiological role [24,25], and Grb2, as a model for other SH2 domain-containing proteins.
The significant activity of compound 1, revealed at 30 µM (Table 2), prompted us to test this compound at different concentrations. Therefore, we calculated its IC50 values against STAT3 and STAT1 to be 15.8 ± 0.6 and ˃ 50 µM, respectively, while the inhibition versus Grb2 was 23% (at a 30 µM concentration).
To identify a trend in terms of structure-activity relationship, we firstly explored the role of the substituents on the benzene ring (1a-i). In this respect, we found that lipophilicity does not play an important role in modulating STAT3 inhibitory activity. Bulky lipophilic substituents at position 5 (derivatives 1g and 1h) induced a significant drop in the inhibition properties (Table 2). Among these derivatives, only 1d, bearing a NO2 group at position 5, exhibited a noticeable activity against STAT3. For this compound, we calculated the IC50 values against STAT3 and STAT1, which were 25.1 ± 2.6 µM and 15.9 ± 2.0 µM, respectively, revealing no selectivity for STAT3 inhibition. Then, we introduced different groups on the nitrogen atoms of the thiadiazole ring (2a,b, 3, and 4), in order to explore asymmetrical substitutions: The presence of the N-substituent did not improve activity. The enhanced bulkiness and hydrophobicity of these substituents compared to compound 1 did not lead to a better interaction. Finally, compound 6 was purchased from commercial sources.

Biological Evaluation
AlphaScreen is an in vitro competitive binding test used to identify compounds that are able to inhibit the binding of proteins containing an SH2 domain to their corresponding phosphopeptides (5-carboxyfluorescein (FITC)-GpYLPQTV for STAT3 and FITC-GpYDKPHVL for STAT1) ( Table 2). Compounds characterized by an interesting STAT3 affinity were further investigated for their selectivity versus STAT1, which shows a high degree of sequence similarity to STAT3 but an opposite physiological role [24,25], and Grb2, as a model for other SH2 domain-containing proteins.
The significant activity of compound 1, revealed at 30 µM (Table 2), prompted us to test this compound at different concentrations. Therefore, we calculated its IC 50 values against STAT3 and STAT1 to be 15.8 ± 0.6 and >50 µM, respectively, while the inhibition versus Grb2 was 23% (at a 30 µM concentration).
To identify a trend in terms of structure-activity relationship, we firstly explored the role of the substituents on the benzene ring (1a-i). In this respect, we found that lipophilicity does not play an important role in modulating STAT3 inhibitory activity. Bulky lipophilic substituents at position 5 (derivatives 1g and 1h) induced a significant drop in the inhibition properties (Table 2). Among these derivatives, only 1d, bearing a NO 2 group at position 5, exhibited a noticeable activity against STAT3. For this compound, we calculated the IC 50 values against STAT3 and STAT1, which were 25.1 ± 2.6 µM and 15.9 ± 2.0 µM, respectively, revealing no selectivity for STAT3 inhibition. Then, we introduced different groups on the nitrogen atoms of the thiadiazole ring (2a,b, 3, and 4), in order to explore asymmetrical substitutions: The presence of the N-substituent did not improve activity. The enhanced bulkiness and hydrophobicity of these substituents compared to compound 1 did not lead to a better interaction.  of compounds 1, 1a-i, 2a-b, and 3 Finally, we investigated the importance of the thiadiazole ring (5 and 6) through the exploration of the function of the SO2 group and of the nitrogen atoms: The results of the AlphaScreen-based assay indicated that these moieties play a key role in the interaction with the STAT3-SH2 domain.
Overall, our data seem to indicate that this class of benzothiadiazole derivatives is characterized by an extremely "tight" SAR: Most of the modifications applied to the scaffold of compound 1 resulted in a considerable loss of activity, with the sole exception of 1d, which retained a partial inhibitory effect. Table 2. Inhibitory activities of compounds 1, 1a-i, 2a-b, and 3 Since from the literature it is known that some benzothiadiazoles could covalently bind to proteins [26], an in silico prediction to evaluate whether the proposed derivatives were pan-assay interference compounds (PAINS) was performed by submitting their structures to five on-line services (FAF-Drugs4 [27], PAINS remove [28], SmartsFilter, (http://pasilla.health.unm.edu/tomcat/biocomp/smartsfilter), SwissADME [29], and Zinc Patterns (http://zinc15.docking.org/patterns/home/)). Three of them (SmartsFilter, SwissADME, and Zinc Patterns) identified compounds 1, 1a-i, and 2a as potential PAINS (see Table S6).
This result was experimentally confuted, at least for compound 1, because it was found to act as a selective STAT3 inhibitor with an IC50 of 15.8 ± 0.6 µM, which was significantly lower than the IC50 shown versus STAT1 (˃50 µM). This selectivity was remarkable considering the high similarity and identity of both STAT1 and STAT3, for the SH2 domain in particular (73.5% sequence similarity and 56.1% sequence identity, calculated by EMBOSS Needle [30]).
Furthermore, to gain insight into the activity of compound 1, its interaction with the cysteine residues located in the vicinities of the SH2 domain (Figure 3) was investigated and evaluated by an AlphaScreen-based assay using different mutants (Table 3).

Compound
Substituents STAT3 % Inhibition (30 µM) [a] STAT3 inhibition: IC 50  Finally, we investigated the importance of the thiadiazole ring (5 and 6) through the exploration of the function of the SO 2 group and of the nitrogen atoms: The results of the AlphaScreen-based assay indicated that these moieties play a key role in the interaction with the STAT3-SH2 domain.
Overall, our data seem to indicate that this class of benzothiadiazole derivatives is characterized by an extremely "tight" SAR: Most of the modifications applied to the scaffold of compound 1 resulted in a considerable loss of activity, with the sole exception of 1d, which retained a partial inhibitory effect.
This result was experimentally confuted, at least for compound 1, because it was found to act as a selective STAT3 inhibitor with an IC 50 of 15.8 ± 0.6 µM, which was significantly lower than the IC 50 shown versus STAT1 ( >50 µM). This selectivity was remarkable considering the high similarity and identity of both STAT1 and STAT3, for the SH2 domain in particular (73.5% sequence similarity and 56.1% sequence identity, calculated by EMBOSS Needle [30]).
Furthermore, to gain insight into the activity of compound 1, its interaction with the cysteine residues located in the vicinities of the SH2 domain (Figure 3) was investigated and evaluated by an AlphaScreen-based assay using different mutants (Table 3).   Compound 1 was inactivated by the addition of an excess amount of cysteine into the assay medium (see Table S7). Moreover, the point mutations of Cys468, Cys542, Cys550, Cys687, or Cys712 in STAT3 conferred resistance to the compound (see Table S8), suggesting that each cysteine residue partially contributes to the inhibition mediated by the molecule.
Though STAT3 contains twelve cysteine residues, the assays were performed only on clones bearing mutations on the five cysteines located near the STAT3-SH2 domain. Therefore, we cannot exclude the possibility that other cysteines may be involved in the interaction with compound 1.
Considering the affinity of compound 1 for STAT3 wild-type, we decided to explore its mechanism of interaction with cysteines by means of analytical studies.
In order to analyze the binding mode of compound 1 to the pockets containing the mutated cysteine residues, a molecular docking study was carried out. In particular, we applied the same protocol used for the virtual screening, but, in this case, the grid maps were calculated by selecting the atoms included in a sphere of 12 Å radius that were centered on each cysteine residue. The so-obtained complexes were optimized with the same minimization protocol employed for the preparation of the STAT3 crystal structure (30,000 steps of conjugate gradients minimization). For a better analysis of the poses, the distances between the cysteine sulfur and the two hypothetical reactive centers (the methyl in 5 and the aromatic carbon in 4) of compound 1 were evaluated. The results in Table S9 show that only the complex with Cys712 seems to have the suitable prerequisites in terms of distances to the reactive centers (4.70 and 4.95 Å for CH3-S and CH-S, respectively; see Figure S72) and docking energy (−5.93 Compound 1 was inactivated by the addition of an excess amount of cysteine into the assay medium (see Table S7). Moreover, the point mutations of Cys468, Cys542, Cys550, Cys687, or Cys712 in STAT3 conferred resistance to the compound (see Table S8), suggesting that each cysteine residue partially contributes to the inhibition mediated by the molecule.
Though STAT3 contains twelve cysteine residues, the assays were performed only on clones bearing mutations on the five cysteines located near the STAT3-SH2 domain. Therefore, we cannot exclude the possibility that other cysteines may be involved in the interaction with compound 1.
Considering the affinity of compound 1 for STAT3 wild-type, we decided to explore its mechanism of interaction with cysteines by means of analytical studies.
In order to analyze the binding mode of compound 1 to the pockets containing the mutated cysteine residues, a molecular docking study was carried out. In particular, we applied the same protocol used for the virtual screening, but, in this case, the grid maps were calculated by selecting the atoms included in a sphere of 12 Å radius that were centered on each cysteine residue. The so-obtained complexes were optimized with the same minimization protocol employed for the preparation of the STAT3 crystal structure (30,000 steps of conjugate gradients minimization). For a better analysis of the poses, the distances between the cysteine sulfur and the two hypothetical reactive centers (the methyl in 5 and the aromatic carbon in 4) of compound 1 were evaluated. The results in Table S9 show that only the complex with Cys712 seems to have the suitable prerequisites in terms of distances to the reactive centers (4.70 and 4.95 Å for CH 3 -S and CH-S, respectively; see Figure S72) and docking energy (−5.93 Kcal/mol), even if the latter parameter must be taken into account with caution because it was calculated by molecular mechanics. The complex with Cys550 also seemed to have the right distances to react with compound 1 (3.64 and 5.13 Å for CH 3 -S and CH-S, respectively); however, the interaction energy was found to be significantly worse than that of the complex with Cys712 (−3.52 versus −5.93 Kcal/mol). Unfortunately, this kind of behavior could not be confirmed by the experimental data; this was probably due to the intrinsic problem afflicting all docking calculations that could only evaluate the physicochemical and steric complementarity of two interacting partners and not the reactivity of the atoms.

MS, UV, and LC Studies
MS studies revealed that compound 1 reacts with glutathione (GSH, used as a model for cysteine-containing peptides) to generate stable covalent adducts. Since no adducts were detectable at the incubation starting point, such reactions did not take place during sample transfer into the MS analyzer. Similar approaches have already been used to identify reactive intermediates by MS studies [31]. The reaction was pH-dependent and required a neutral or basic pH to take place, whereas no reactivity was observed at an acidic pH ( Figure 4). Kcal/mol), even if the latter parameter must be taken into account with caution because it was calculated by molecular mechanics. The complex with Cys550 also seemed to have the right distances to react with compound 1 (3.64 and 5.13 Å for CH3-S and CH-S, respectively); however, the interaction energy was found to be significantly worse than that of the complex with Cys712 (−3.52 versus −5.93 Kcal/mol). Unfortunately, this kind of behavior could not be confirmed by the experimental data; this was probably due to the intrinsic problem afflicting all docking calculations that could only evaluate the physicochemical and steric complementarity of two interacting partners and not the reactivity of the atoms.

MS, UV, and LC Studies
MS studies revealed that compound 1 reacts with glutathione (GSH, used as a model for cysteinecontaining peptides) to generate stable covalent adducts. Since no adducts were detectable at the incubation starting point, such reactions did not take place during sample transfer into the MS analyzer. Similar approaches have already been used to identify reactive intermediates by MS studies [31]. The reaction was pH-dependent and required a neutral or basic pH to take place, whereas no reactivity was observed at an acidic pH ( Figure 4). The absence of reactivity at the acidic pH can be explained on one hand by the pH-dependent nucleophilicity of GSH [32] and on the other hand by the deprotonation of compound 1 occurring at a neutral/basic pH (pKa = 6.88 ± 0.03) [22], which can lead to structure rearrangements that possibly contribute to an increased reactivity towards GSH. However, the methyl groups seem fundamental for the covalent reaction, since no adducts were detectable upon incubation of GSH with compound 1d, which was the second most active compound according to the results reported in Table 2.
At a neutral pH ( Figure 4C), a single adduct was detectable at the reaction endpoint (504.12121 m/z), whereas at a basic pH ( Figure 4D), three signals at 502.10569, 504.12120, and 520.11607 m/z were detected along with the complete oxidation of GSH to its dimer GSSG (i.e., the conversion of the singly charged GSH ion at 308.09106 m/z to the doubly charged GSSG ion at 307.08332 m/z). The absence of reactivity at the acidic pH can be explained on one hand by the pH-dependent nucleophilicity of GSH [32] and on the other hand by the deprotonation of compound 1 occurring at a neutral/basic pH (pKa = 6.88 ± 0.03) [22], which can lead to structure rearrangements that possibly contribute to an increased reactivity towards GSH. However, the methyl groups seem fundamental for the covalent reaction, since no adducts were detectable upon incubation of GSH with compound 1d, which was the second most active compound according to the results reported in Table 2.
At a neutral pH ( Figure 4C), a single adduct was detectable at the reaction endpoint (504.12121 m/z), whereas at a basic pH ( Figure 4D), three signals at 502.10569, 504.12120, and 520.11607 m/z were detected along with the complete oxidation of GSH to its dimer GSSG (i.e., the conversion of the singly charged GSH ion at 308.09106 m/z to the doubly charged GSSG ion at 307.08332 m/z).
The signals at 502.10569, 504.12120, and 520.11607 m/z were all within 1 ppm from the theoretical m/z values expected for singly charged ions of molecules C 18  Interestingly, at pH 10, the adducts were also detected at the incubation starting point if compound 1 was added before GSH into the buffer ( Figure 5A), but not if GSH ( Figure 5B) or Na 2 S 2 O 5 ( Figure 5C) were added into the buffer before compound 1.
Molecules 2020, 25, x FOR PEER REVIEW 11 of 29 The signals at 502.10569, 504.12120, and 520.11607 m/z were all within 1 ppm from the theoretical m/z values expected for singly charged ions of molecules C18H25N5O8S2, C18H23N5O8S2, and C18H25N5O9S2, respectively.
Interestingly, at pH 10, the adducts were also detected at the incubation starting point if compound 1 was added before GSH into the buffer ( Figure 5A), but not if GSH ( Figure 5B) or Na2S2O5 ( Figure  5C) were added into the buffer before compound 1. Figure 5. MS spectra at the reaction starting point for a 1:1 GSH:compound 1 mixture incubated at pH 10 (spectra A and B). Samples were prepared by adding compound 1 (spectrum A) or GSH (spectrum B) as the first reagent. MS spectra at the incubation starting point (spectrum C) and endpoint (spectrum D) for a 1:1 GSH:compound 1 mixture incubated at pH 10 with Na2S2O5 in the buffer.
Since GSH is an antioxidant, we speculated on the involvement of oxidation steps for the formation of adducts. To address such a hypothesis, the experiments were repeated in a buffer containing Na2S2O5, which had similar effect of GSH at the incubation starting point ( Figure 5C) and prevented the formation of adducts at 502.10569 and 520.11607 m/z at the incubation endpoint ( Figure 5D), suggesting that such adducts are generated only in oxidizing conditions.
On the contrary, the adduct at 504.12130 m/z was detectable at the incubation endpoint, despite Na2S2O5 addition ( Figure 5D).
Consistently with MS data, LC-UV experiments detected a single adduct when compound 1 was incubated with GSH and Na2S2O5 (i.e., chromatographic peak at RT = 8.7 min, Figure 6B), whereas multiple peaks were generated if no antioxidants were added into the buffer ( Figure 6C). Figure 5. MS spectra at the reaction starting point for a 1:1 GSH:compound 1 mixture incubated at pH 10 (spectra A and B). Samples were prepared by adding compound 1 (spectrum A) or GSH (spectrum B) as the first reagent. MS spectra at the incubation starting point (spectrum C) and endpoint (spectrum D) for a 1:1 GSH:compound 1 mixture incubated at pH 10 with Na 2 S 2 O 5 in the buffer.
Since GSH is an antioxidant, we speculated on the involvement of oxidation steps for the formation of adducts. To address such a hypothesis, the experiments were repeated in a buffer containing Na 2 S 2 O 5 , which had similar effect of GSH at the incubation starting point ( Figure 5C) and prevented the formation of adducts at 502.10569 and 520.11607 m/z at the incubation endpoint ( Figure 5D), suggesting that such adducts are generated only in oxidizing conditions.
On the contrary, the adduct at 504.12130 m/z was detectable at the incubation endpoint, despite Na 2 S 2 O 5 addition ( Figure 5D).
Consistently with MS data, LC-UV experiments detected a single adduct when compound 1 was incubated with GSH and Na 2 S 2 O 5 (i.e., chromatographic peak at RT = 8.7 min, Figure 6B), whereas multiple peaks were generated if no antioxidants were added into the buffer ( Figure 6C). Na 2 S 2 O 5 also prevented the complete consumption of compound 1 (Figure 5B), thus confirming that oxidizing conditions trigger a faster reaction rate.
Moreover, despite MS experiments detecting an adduct at 504.12125 ± 0.00006 m/z, both with and without Na 2 S 2 O 5 , the corresponding chromatograms had no common peaks. This means that oxidizing and non-oxidizing conditions lead to the formation of adducts with different structures but the same elemental composition.
This hypothesis was also supported by the UV spectra collected for the corresponding reaction batches. As reported in Figure 6, the UV spectra at the reaction endpoint look alike for reactions buffered at pH 7.4 or pH 10 with Na 2 S 2 O 5 , with no shifts compared to the starting point (data not shown). On the contrary, the incubation at pH 10 without Na 2 S 2 O 5 was only found to have a similar shape at the starting point, while a significant shift was observed at the reaction endpoint (Figure 7). Molecules 2020, 25, x FOR PEER REVIEW 12 of 29 Figure 6. Chromatograms of GSH incubated with compound 1 as recorded at incubation starting point (A), at incubation endpoint in a pH 10 buffer containing Na2S2O5 (B), and at incubation endpoint in a Na2S2O5-free pH 10 buffer (C).
Na2S2O5 also prevented the complete consumption of compound 1 (Figure 5B), thus confirming that oxidizing conditions trigger a faster reaction rate.
Moreover, despite MS experiments detecting an adduct at 504.12125 ± 0.00006 m/z, both with and without Na2S2O5, the corresponding chromatograms had no common peaks. This means that oxidizing and non-oxidizing conditions lead to the formation of adducts with different structures but the same elemental composition.
This hypothesis was also supported by the UV spectra collected for the corresponding reaction batches. As reported in Figure 6, the UV spectra at the reaction endpoint look alike for reactions buffered at pH 7.4 or pH 10 with Na2S2O5, with no shifts compared to the starting point (data not shown). On the contrary, the incubation at pH 10 without Na2S2O5 was only found to have a similar shape at the starting point, while a significant shift was observed at the reaction endpoint ( Figure 7). The reaction mechanism in Figure 8 accounts for all the observations described above. Specifically, the overall mechanism requires a neutral-to-basic pH for compound 1 deprotonation to generate adducts. All included oxidation steps are dependent on the basic pH activation of compound  Na2S2O5 also prevented the complete consumption of compound 1 (Figure 5B), thus confirming that oxidizing conditions trigger a faster reaction rate.
Moreover, despite MS experiments detecting an adduct at 504.12125 ± 0.00006 m/z, both with and without Na2S2O5, the corresponding chromatograms had no common peaks. This means that oxidizing and non-oxidizing conditions lead to the formation of adducts with different structures but the same elemental composition.
This hypothesis was also supported by the UV spectra collected for the corresponding reaction batches. As reported in Figure 6, the UV spectra at the reaction endpoint look alike for reactions buffered at pH 7.4 or pH 10 with Na2S2O5, with no shifts compared to the starting point (data not shown). On the contrary, the incubation at pH 10 without Na2S2O5 was only found to have a similar shape at the starting point, while a significant shift was observed at the reaction endpoint ( Figure 7). The reaction mechanism in Figure 8 accounts for all the observations described above. Specifically, the overall mechanism requires a neutral-to-basic pH for compound 1 deprotonation to generate adducts. All included oxidation steps are dependent on the basic pH activation of compound The reaction mechanism in Figure 8 accounts for all the observations described above. Specifically, the overall mechanism requires a neutral-to-basic pH for compound 1 deprotonation to generate adducts. All included oxidation steps are dependent on the basic pH activation of compound 1, like with catechol [33], and provide the generation of reactive intermediates similar to quinones and quinone-methides that are able to covalently react with GSH [34][35][36]. The final products VIb and VIIa are generated only after the oxidative rearrangement of compound 1 and have molecular formulas in agreement with the MS signals at 502 and 520 m/z, found only in reaction batches without antioxidants. The products IVa and IIb were found to have the same molecular formula, which was in agreement with the signal at 504 m/z that was found in reaction batches with or without antioxidants. However, consistently with LC experiments, the products IVa and IIb were found to have different structures and formation mechanisms: IVa was found to be generated after compound 1 oxidation, while the formation of IIb was not found to require the oxidation of compound 1. Interestingly, such mechanism is not specific for the class of compounds since it was only observed for compound 1. The removal of the methyl groups led to compounds that were unable to covalently react with glutathione, despite some of them still being capable of inhibiting STAT3 activity (e.g., compound 1d; see Table 2). generated only after the oxidative rearrangement of compound 1 and have molecular formulas in agreement with the MS signals at 502 and 520 m/z, found only in reaction batches without antioxidants. The products IVa and IIb were found to have the same molecular formula, which was in agreement with the signal at 504 m/z that was found in reaction batches with or without antioxidants. However, consistently with LC experiments, the products IVa and IIb were found to have different structures and formation mechanisms: IVa was found to be generated after compound 1 oxidation, while the formation of IIb was not found to require the oxidation of compound 1. Interestingly, such mechanism is not specific for the class of compounds since it was only observed for compound 1. The removal of the methyl groups led to compounds that were unable to covalently react with glutathione, despite some of them still being capable of inhibiting STAT3 activity (e.g., compound 1d; see Table 2).

1 H-NMR Analysis
With the aim of confirming the stated hypotheses, NMR studies were performed in the experimental conditions previously considered for the MS analysis in the absence or in the presence of GSH. Actually, few NMR studies concerning covalent interactions between GSH/STAT proteins and small molecules have been reported. In one case [37], the authors exploited the presence of fluorine atoms on the studied compounds and used 19 F-NMR spectroscopy, which has the advantage of simple spectra with a wide spread of resonances.
In our case, we used 1 H-NMR spectroscopy even though it is known that GSH, like other small peptides, presents a great conformational variability, also depending on the pH, that influences its ionization state [38]; consequently, its 1 H-NMR spectra could appear quite complex.
Therefore, to study the interaction of compound 1 with the model peptide GSH, we decided to perform a comparison, at different times, of the spectra of compound 1 and of GSH alone with the spectrum of a mixture of them at pH 10 and 37 °C. These conditions were chosen because they proved

1 H-NMR Analysis
With the aim of confirming the stated hypotheses, NMR studies were performed in the experimental conditions previously considered for the MS analysis in the absence or in the presence of GSH. Actually, few NMR studies concerning covalent interactions between GSH/STAT proteins and small molecules have been reported. In one case [37], the authors exploited the presence of fluorine atoms on the studied compounds and used 19 F-NMR spectroscopy, which has the advantage of simple spectra with a wide spread of resonances.
In our case, we used 1 H-NMR spectroscopy even though it is known that GSH, like other small peptides, presents a great conformational variability, also depending on the pH, that influences its ionization state [38]; consequently, its 1 H-NMR spectra could appear quite complex.
Therefore, to study the interaction of compound 1 with the model peptide GSH, we decided to perform a comparison, at different times, of the spectra of compound 1 and of GSH alone with the spectrum of a mixture of them at pH 10 and 37 • C. These conditions were chosen because they proved to be optimal for the interaction between the two compounds in the MS investigations. Moreover, in order to make the NMR study easier, we decided to add Na 2 S 2 O 5 to the mixture to reduce the number of adducts that could potentially derive from the reactions between compound 1 and GSH. The use of this reducing agent made the MS spectra less complex, mainly showing a 504 [M + H] + ion ( Figure 5D) corresponding to few hypothetical structures (IIb and IVa; Figure 8); in the same conditions, the LC chromatogram revealed essentially one peak at 8.7 min, even if this analysis showed only a reduced transformation of the starting compound 1 at the incubation endpoint ( Figure 6B).
Moreover, the addition of Na 2 S 2 O 5 should have prevented the presence of interfering NMR resonances due to the formation of oxidized GSSG.
Though this study did not demonstrate the formation of the expected compounds (see discussion in Supplementary Materials and Figures S72-S74), we believe that the NMR outcomes cannot disprove the results obtained by MS spectroscopy.

Computational Methods
The structural coordinates of STAT3, co-crystallized with a DNA fragment, were downloaded from the Protein Data Bank (PDB ID 1BG1) [19]. The protein was dimerized by applying the transformation matrix, as shown in the PDB file, and the model was completed by the addition of the hydrogens in two steps: (1) to STAT3, applying the algorithm for proteins, and (2) to DNA, applying the algorithm for nucleic acids. In both cases, we used the features included in the VEGA ZZ package [39]. Atom charges (Gasteiger-Marsili method) [40] and potentials (CHARMM 22 for proteins and nucleic acids) [41,42] were assigned to the obtained structure. Finally, the model was optimized by a conjugate gradient minimization (30,000 steps) to reduce the high-energy steric interactions. In order to preserve the experimental data, atom constraints were applied to protein and DNA backbones. This step was carried out by NAMD 2.13 [43], which was integrated in the VEGA ZZ graphic environment. To perform virtual screening calculations, the following databases were chosen: AKos (AKos Consulting & Solutions GmbH, Steinen, Germany; 544,391 molecules), ChemPDB (4009), ChemBank (Broad Institute, Cambridge, MA, USA; 2344), KEGG (Kyoto Encyclopedia of Genes and Genomes, Kanehisa Laboratories, Kyoto, Japan; 10005), NCI's Anti-HIV (National Cancer Institute, National Institute of Health, USA; 42689), and NCI (National Cancer Institute, National Institute of Health, USA; 15237). These databases are available for free at the mirror site of Ligand.info (http://biophysics.med.jhmi.edu/~{}yliu120/dockingdatabase.html; the original database site is no longer available) [44], which includes collections of compounds for virtual screening. These structures are already converted to 3D and fully optimized by molecular mechanics. Nevertheless, all molecules were pre-processed setting their ionization state at a physiological pH, and the resulting structures were optimized and refined by PM7 semi-empirical method, as implemented in MOPAC 2016. To perform this calculation, the WarpEngine technology implemented in VEGA ZZ [45] was used, reducing the computational time to few tens of minutes for the whole dataset through the distribution of the MOPAC calculation on a network of PCs. In more detail: Three blade servers equipped with two CPUs, each of Intel Xeon E5 class, were used for a total of 52 physical cores and 104 threads. The virtual screening was carried out by the GriDock software, which was especially designed to run on high performance computing systems (HPC) and distributed computer resources (GRID) as front-end to the well-known AutoDock 4 package [20]. Specifically, GriDock is able to extract each molecule from a database and perform a docking calculation through AutoDock 4. Though it is currently possible to perform screenings with faster and more efficient programs, we preferred to spend more time on the calculation because we had already obtained good results with AutoDock in previous studies on STAT3. Before running GriDock, a single STAT3 monomer was considered, and the grid maps, required to evaluate docking scores, were calculated by selecting the atoms included in a sphere of 12 Å radius centered on the phosphorylated Tyr-705 (PTR-705 in the PDB file). This pocket represented a possible hot spot of interaction with STAT3 for ligands, since it is known to play a pivotal role in STAT3 dimerization and activation. This process was carried out by AutoGrid 4 interfaced to VEGA ZZ. The compounds of the considered databases were docked by GriDock/AutoDock using the genetic-algorithm search and generating ten possible solutions.

Materials and Methods
All starting materials, chemicals, and solvents were purchased from commercial suppliers (Sigma-Aldrich-Merck, St. Louis, MI, USA; FluoroChem, Hadfield, UK) and used as received. Anhydrous solvents were utilized without further drying. Aluminum-backed silica gel 60 plates Procedure E: To a stirred solution of 1d (140 mg, 0.65 mmol) in dry DMF (2 mL), dried K 2 CO 3 (179.7 mg, 1.3 mmol) and PMB-Br (313.6 mg, 1.56 mmol) were added. The mixture was stirred at 80 • C for 2 h and then cooled to room temperature, diluted with H 2 O (6 mL), and extracted with EtOAc (2 × 5 mL). The collected organic layers were dried over anhydrous Na 2 SO 4 , filtered, and evaporated in vacuo. The crude residue was purified by flash chromatography ( Intermediate 14 (114.9 mg, 0.27 mmol) was dissolved in acetone (2 mL), and the mixture was cooled to 0 • C in an ice bath. Acetic anhydride (41.3 mg, 0.41 mmol) was added dropwise, and the solution was stirred at room temperature for 2 h. After completion, the solvent was removed in vacuo and the residue was diluted with 1 M NaOH and extracted with EtOAc (2 × 3 mL). The collected organic layers were dried over anhydrous Na 2 SO 4 , filtered, and concentrated in vacuo. The crude residue was purified by flash column chromatography (cyclohexane/EtOAc

Conclusions
Benzothiadiazole derivative 1 was identified as a novel STAT3 inhibitor by structure-based virtual screening performed on the SH2 domain. The AlphaScreen-based assay confirmed, albeit indirectly, that compound 1 was able to inhibit STAT3 dimerization, exhibiting a good activity (STAT3 IC 50 = 15.8 ± 0.6 µM versus STAT1 IC 50 > 50 µM). Therefore, a small set of derivatives was designed and synthesized through a flexible diversity-oriented synthetic approach when required. The compounds were tested to define the main structural features for the lead development. Among them, only compound 1d exhibited a significant, albeit reduced, activity, thus suggesting a considerably "tight" SAR for this class of derivatives. Moreover, because compound 1, like other known STAT3 inhibitors, probably binds to cysteine residues, MS, UV, LC, and NMR studies were performed to understand its mechanism of interaction. Mimicking the biological conditions, we conducted an MS analysis of compound 1 in the presence of GSH: We discovered that this inhibitor can covalently bind to cysteine residues through a Michael-like reaction, leading to the formation of some alkylation adducts.
In an attempt to confirm the results of the MS study, an NMR investigation was carried out to further characterize the adducts formed by the interaction of compound 1 with GSH. However, despite the same reaction conditions (pH, temperature, and use of an antioxidant) being employed, the intrinsic differences between the two techniques prevented us from obtaining additional evidence to support the MS conclusions. Notably, NMR spectroscopy is seldom used for this kind of study in the literature [16,18], which suggests that its inherent limitations make it unsuitable for such an application. Therefore, it would seem reasonable to claim that the NMR results neither supported nor disproved our hypotheses.
The research of STAT3 inhibitors has led to several potent anti-tumor candidates over the years (e.g., Stattic, S3I-201, and erasin). In this context, we are confident that our ongoing studies will allow us to develop and improve our lead compound 1 as a new PPI inhibitor endowed with a novel mechanism of interaction and characterized by a pH-dependent activation.
Supplementary Materials: The following are available online: Tables S1-S5: Top-ranked compounds from computational studies; Figure S1: Main SH2 domain residues involved in the interaction with compound 1; Table S6: Pan-assay interference compound (PAINS) in silico prediction test, employing five on-line services (SmartsFilter, SwissADME, Zinc Patterns, FAF-Drugs4, and PAINS remover); Figures S2-S70: 1 H and 13 C-NMR spectra of synthesized compounds; Table S7: Effect of cysteine on the inhibitory activity of compound 1 versus STAT3; Table S8: Effect of the mutation of cysteine residues located in the vicinities of the SH2 domain on the inhibitory activity of compound 1 versus STAT3; Figure S71: CBB staining of the STAT3(136-705) and STAT5b(136-703) proteins. The purified soluble proteins (0.5 µg) were analyzed by SDS-PAGE; Table S9: Comparison of the best STAT3 complexes obtained by docking compound 1 into the pockets containing the mutated cysteines, considered in the mutagenesis study; Figure