Computational Methods in Cooperation with Experimental Approaches to Design Protein Tyrosine Phosphatase 1B Inhibitors in Type 2 Diabetes Drug Design: A Review of the Achievements of This Century

Protein tyrosine phosphatase 1B (PTP1B) dephosphorylates phosphotyrosine residues and is an important regulator of several signaling pathways, such as insulin, leptin, and the ErbB signaling network, among others. Therefore, this enzyme is considered an attractive target to design new drugs against type 2 diabetes, obesity, and cancer. To date, a wide variety of PTP1B inhibitors that have been developed by experimental and computational approaches. In this review, we summarize the achievements with respect to PTP1B inhibitors discovered by applying computer-assisted drug design methodologies (virtual screening, molecular docking, pharmacophore modeling, and quantitative structure–activity relationships (QSAR)) as the principal strategy, in cooperation with experimental approaches, covering articles published from the beginning of the century until the time this review was submitted, with a focus on studies conducted with the aim of discovering new drugs against type 2 diabetes. This review encourages the use of computational techniques and includes helpful information that increases the knowledge generated to date about PTP1B inhibition, with a positive impact on the route toward obtaining a new drug against type 2 diabetes with PTP1B as a molecular target.


Introduction
Cancer, obesity, and diabetes mellitus are major health problems worldwide, causing around 9.6, 2.8, and 1.6 million deaths a year, respectively [1]. In addition, all of these disease impose a high expense on the health system and even more when they coexist [2]. Despite the wide variety of oral and injectable therapies available for these diseases, their uses are limited by efficacy, adverse effects, and contraindications [3][4][5][6][7]. Therefore, the increased prevalence of these diseases highlights the necessity of searching for new drugs for their treatment.
In this sense, protein tyrosine phosphatase 1B (PTP1B) has been established as a pharmacological target for these pathologies. Its substrates are involved in multiple cellular processes, such as glucose homeostasis regulated by insulin signaling, decreased food intake, increased energy expenditure, cellular proliferation, and more. PTP1B is considered an interesting drug target for the treatment of type 2 diabetes because it dephosphorylates the insulin receptor (IR) and the insulin receptor substrate 1 (IRS-1), inactivating the downstream pathway of phosphatidylinositol 3-kinase (PI3K)-Akt and preventing the translocation of glucose transporter 4 (GLUT4) [8,9] (Figure 1). In addition, attention has been on PTP1B as a potentially excellent therapeutic target in obesity, owing to the role that it plays in the regulation of leptin signaling. PTP1B downregulates leptin receptor (LR) stimuli by disrupting the autophosphorylation of JAK2 and subsequent LR phosphorylation, as well as the activation of STAT3, which mediates the transcription of target genes [10,11] ( Figure 1). On the other hand, it has been reported that PTP1B can block the induction of cell proliferation and cell survival via the negative regulation of the signaling cascade linking ErbB2-PTP1B-Src kinase, which contributes to the tumor-suppressor function of PTP1B in cancer cells [12][13][14] (Figure 1). However, this last role of PTP1B has become increasingly controversial; for an extensive review, the reader is referred to [15,16]. The above-mentioned controversy highlights the relevance of developing therapies targeting PTP1B in these diseases. food intake, increased energy expenditure, cellular proliferation, and more. PTP1B is considered an interesting drug target for the treatment of type 2 diabetes because it dephosphorylates the insulin receptor (IR) and the insulin receptor substrate 1 (IRS-1), inactivating the downstream pathway of phosphatidylinositol 3-kinase (PI3K)-Akt and preventing the translocation of glucose transporter 4 (GLUT4) [8,9] (Figure 1). In addition, attention has been on PTP1B as a potentially excellent therapeutic target in obesity, owing to the role that it plays in the regulation of leptin signaling. PTP1B downregulates leptin receptor (LR) stimuli by disrupting the autophosphorylation of JAK2 and subsequent LR phosphorylation, as well as the activation of STAT3, which mediates the transcription of target genes [10,11] (Figure 1). On the other hand, it has been reported that PTP1B can block the induction of cell proliferation and cell survival via the negative regulation of the signaling cascade linking ErbB2-PTP1B-Src kinase, which contributes to the tumor-suppressor function of PTP1B in cancer cells [12][13][14] (Figure 1). However, this last role of PTP1B has become increasingly controversial; for an extensive review, the reader is referred to [15,16]. The above-mentioned controversy highlights the relevance of developing therapies targeting PTP1B in these diseases. Accordingly, the development of PTP1B inhibitors has increased considerably for the treatment of these diseases. Nevertheless, PTP1B inhibitors have not progressed beyond the preclinical stage due to bioavailability and specificity challenges [17,18]. One of the principal hindrances is the chemical environment of the PTP1B active site, which is highly polar; therefore, it attracts negatively charged molecules with poor membrane permeability and limited oral bioavailability [19]. Regarding selectivity, the PTP family is characterized by an exceptionally high degree of sequence conservation across active sites [20]. Tcell protein tyrosine phosphatase (TCPTP) is the closest homolog of PTP1B [21,22]. It has been reported that mice lacking TCPTP die 5 weeks after birth due to defects in immune function and hematopoiesis failure [23]. In contrast, PTP1B knockout mice were reported to have a normal life span [24]. Therefore, it is important to consider these challenges in the design of PTP1B inhibitors.
Moreover, drug development is a costly and complex process that consumes a lot of time-around 10-15 years [25,26]. Recently, computer-assisted drug design (CADD) has Accordingly, the development of PTP1B inhibitors has increased considerably for the treatment of these diseases. Nevertheless, PTP1B inhibitors have not progressed beyond the preclinical stage due to bioavailability and specificity challenges [17,18]. One of the principal hindrances is the chemical environment of the PTP1B active site, which is highly polar; therefore, it attracts negatively charged molecules with poor membrane permeability and limited oral bioavailability [19]. Regarding selectivity, the PTP family is characterized by an exceptionally high degree of sequence conservation across active sites [20]. T-cell protein tyrosine phosphatase (TCPTP) is the closest homolog of PTP1B [21,22]. It has been reported that mice lacking TCPTP die 5 weeks after birth due to defects in immune function and hematopoiesis failure [23]. In contrast, PTP1B knockout mice were reported to have a normal life span [24]. Therefore, it is important to consider these challenges in the design of PTP1B inhibitors.
Moreover, drug development is a costly and complex process that consumes a lot of time-around 10-15 years [25,26]. Recently, computer-assisted drug design (CADD) has become a crucial component used by the pharmaceutical industry and academic institutions to decrease costs and is a huge time saver due to the minimization of laboratory work and experimentation. With this review, we aim to present the efforts made to obtain PTP1B inhibitors through CADD, with particular focus on the protein sites used to this end, as well as the computational strategy, supported by the inhibitory activity evaluated in vitro. Our review included only articles published during this century and focused on PTP1B as a target for type 2 diabetes drug design. The research was carried out using various groups of keywords, such as PTP1B inhibitors, virtual screening, molecular docking, pharmacophore modeling, 3D-QSAR, and computational drug design.

PTP1B Structure and Target Sites
PTP1B, a ubiquitously expressed classical non-receptor PTP, is encoded by the PTPN1 gene. This enzyme hydrolyzes phosphotyrosine (pTyr)-containing proteins. In the cell, it is a protein of~50 kDa (435 amino acids) located in the endoplasmic reticulum; however, it was originally isolated from a human placenta as a 37 kDa protein that includes the catalytic domain (residues 1-321) [27][28][29]. Structurally, the enzyme is formed by an N-terminal catalytic domain, two proline-rich sequences, and a C-terminal hydrophobic region (35 residues), which serves to attach the enzyme to the membrane of the endoplasmic reticulum [28,30,31]. Furthermore, it contains a regulatory segment of~115 residues on the C-terminal side of the catalytic domain [27] (Figure 2). PTP1B inhibitors through CADD, with particular focus on the protein sites used to this end, as well as the computational strategy, supported by the inhibitory activity evaluated in vitro. Our review included only articles published during this century and focused on PTP1B as a target for type 2 diabetes drug design. The research was carried out using various groups of keywords, such as PTP1B inhibitors, virtual screening, molecular docking, pharmacophore modeling, 3D-QSAR, and computational drug design.

PTP1B Structure and Target Sites
PTP1B, a ubiquitously expressed classical non-receptor PTP, is encoded by the PTPN1 gene. This enzyme hydrolyzes phosphotyrosine (pTyr)-containing proteins. In the cell, it is a protein of ~50 kDa (435 amino acids) located in the endoplasmic reticulum; however, it was originally isolated from a human placenta as a 37 kDa protein that includes the catalytic domain (residues 1-321) [27][28][29]. Structurally, the enzyme is formed by an N-terminal catalytic domain, two proline-rich sequences, and a C-terminal hydrophobic region (35 residues), which serves to attach the enzyme to the membrane of the endoplasmic reticulum [28,30,31]. Furthermore, it contains a regulatory segment of ~115 residues on the C-terminal side of the catalytic domain [27] (Figure 2).
PTP1B was the first PTP structure reported at high resolution [32]. To date, 274 human PTP1B structures have been deposited in the Protein Data Bank (PDB, www.rcsb.org, accessed on 1 February 2022). Of these, around two hundred twenty-five structures are in complex with different ligands: nine in the apo form and the rest with some mutations or chemical modifications, all with a resolution ranging from 1.5 to 3.3 Å. Most crystallographic complexes are formed with competitive inhibitors bound at the PTP1B active site. Nevertheless, there are some complexes with inhibitors bound at distinct sites, such as the C-terminal region of the catalytic domain (helix α9) [33] and at a site formed by helices α3, α6, and α7 called an allosteric site [34] (Figure 2).  PTP1B was the first PTP structure reported at high resolution [32]. To date, 274 human PTP1B structures have been deposited in the Protein Data Bank (PDB, www.rcsb.org, accessed on 1 February 2022). Of these, around two hundred twenty-five structures are in complex with different ligands: nine in the apo form and the rest with some mutations or chemical modifications, all with a resolution ranging from 1.5 to 3.3 Å. Most crystallographic complexes are formed with competitive inhibitors bound at the PTP1B active site. Nevertheless, there are some complexes with inhibitors bound at distinct sites, such as the C-terminal region of the catalytic domain (helix α9) [33] and at a site formed by helices α3, α6, and α7 called an allosteric site [34] (Figure 2).

Active and Secondary Sites
The PTP1B active site is formed by several loop regions (P-, WPD-, Q-, E-, and pTyr loops) and a secondary site ( Figure 2). The phosphate-binding loop, i.e., the P loop or PTP loop (residues 214-221), contains the conserved signature motif VHCSXGXGR[T/S]G, including the catalytic Cys215 and the invariant Arg221, which provides specificity for pTyr in classical PTPs [20,35]. The WPD loop (residues 179 to 189) acts as a flexible gate to the active site that can take closed (active) and open (inactive) conformations, where substrate binding only occurs when the loop is in the open state [35,36]. Another loop is the Q loop (residues 262 to 266), which contains conserved Glu262 that coordinates the water molecule necessary for hydrolysis of the thiophosphate intermediate [35]. The E loop contains multiple conserved residues; it has been suggested that this loop coordinates the dynamics of the WPD loop [37]. The most important residue is the conserved Lys120, which interacts with the catalytic aspartate (Asp181) of the WPD loop (in its closed conformation) by a hydrogen bond; this interaction stabilizes substrate-bound conformation [38]. Finally, the pTyr loop, which is present in all classical protein tyrosine phosphatases, is considered a pTyr recognition loop and is responsible for the selectivity of pTyr over pSer/pThr [37]. It is characterized by the sequence NXXKNRY, where Tyr46 recognizes the pTyr residue of the substrate and facilitates its access to the active site through electrostatic interactions, Asn44 strengthens the interaction between this residue and active-site residues by forming hydrogen bonds, and Arg45 stabilizes the loop [35,39,40]. All these conserved loops structures are relevant for phosphatase activity and are required for several activities, such as substrate recognition, binding, and catalysis [41].
On the other hand, an additional pocket has been described, called the second aryl phosphate-binding site, B site, or secondary site. This is an important non-conserved site that regulates the substrate specificity [42,43]. It was discovered through an unexpected binding mode observed in the complex with the compound bis-(para-phosphophenyl) methane [43]. The most important residues in this site are Arg24 and Arg254; their guanidinium moieties interact with oxygen atoms of the proximal phosphate and participate in the recognition of diphosphorylated substrates [42].

Allosteric Sites
As previously mentioned, two allosteric sites have been reported in PTP1B: the C-terminal segment (helix α9, residues 367-394) [33] and a site formed by helices α3, α6, and α7 [34]. The first site was identified using a long form of PTP1B (residues 1-393) in a complex with compound MSI-1436 (trodusquemine). NMR studies indicated that the trodusquemine primary binding site is located between residues 367 and 394 in helix α9 [33].
The other site was discovered by crystalizing PTP1B in complex with benzofuran derivatives. This site is a hydrophobic pocket formed by the side chains of Leu192, Phe196, and Phe280 and is located between the α7, α3, and α6 helices. It was observed that when inhibitors interacted with these helices, especially α7, closure of the WPD loop was prevented [34].

Computational Strategies Applied to Discover PTP1B Inhibitors
A considerable number of PTP1B inhibitors have been reported in recent decades. Many have not only been discovered by experimental procedures but also by using computational methods. Such methods have allowed for the successful development of selective and potent PTP1B inhibitors, in addition to guiding the optimization of different scaffolds [44][45][46][47][48][49]. Molecular docking, virtual screening, pharmacophore modeling, and QSAR are some of the strategies used to this end and will be briefly described hereafter.

Virtual Screening
Virtual screening is a strategy that permits the search for molecules with potential biological activity in large chemical libraries using a computational model [50,51]. Many computational models can be employed to perform virtual screening analysis. They can be categorized as ligand-based and receptor-based virtual screening. Ligand-based methods, for which no information on the receptor is necessary, leverage the information provided by a compound or a set of compounds that is active on the desired target to identify other molecules in the database with similar structural characteristics [52]. On the other hand, receptor-based strategies require knowledge of the 3D structure of the target receptor binding site to select compounds according to their likelihood to bind to the receptor. These involve molecular docking of each ligand to the binding site of the target. In both categories, the obtained information is then used to rank the compounds to select and experimentally evaluate a small subset for biological activity [53]. In addition, the virtual screening approach allows for screening of molecules that do not necessarily exist physically but which can be obtained through purchase or synthesis, enriching ligand libraries and saving time and money [54]. However, this computational approach is incapable of correctly ranking all the molecules in a library or finding all possible active compounds due to the inaccuracy of the scoring functions employed to identify active molecules [55]. The reader is referred to [56][57][58][59] for a comprehensive review.

Molecular Docking
Molecular docking predicts the binding mode of chemical entities within the targeting cavity of the receptor of interest and provides an estimated binding affinity value through a search algorithm and energy-scoring function [60][61][62]. The common steps to carry out this approach are as follows: selection of target and ligand 3D structures, followed by preparation of those structures, depending on the requirements of the molecular docking protocol being employed [63,64]. Three types of docking can be chosen: rigid, semiflexible, or flexible docking. For rigid docking, both the receptor and ligand are kept in a rigid position [61]. Semi-flexible involves a motionless receptor, whereas ligand flexibility is permitted, which allows for a faster and more direct process [64]. Flexible docking is much more computationally intensive than rigid docking, owing to flexible ligands and receptors [65]. Additionally, the results obtained with this tool can be improved by combining different scoring functions to obtain a consensus score [66]. On the other hand, other scoring functions allow for establishment of a new molecule by screening small compounds (fragments) in a receptor cavity; this is called fragment-based docking. This strategy helps to optimize hits, improving the interaction at the binding site, as well as their physicochemical, pharmacokinetic, and toxicological properties, and new chemical libraries are generated to be synthesized [67]. The reader is referred to [68][69][70] for a comprehensive review.

Pharmacophore Modeling
Another strategy to select molecules from large libraries is pharmacophore modeling, which allows for selection of compounds with similar chemical and physical features, assuming that they have related biological activity [71,72]. A pharmacophore model can be built by superposing a set of active molecules and extracting common chemical features that are necessary for their biological activity; this process is called ligand-based pharmacophore modeling [44]. Moreover, a structure-based pharmacophore modeling strategy can be followed. This consists of obtaining the essential chemical features for the optimal interaction between a biological receptor and a ligand; however, it is necessary to know the receptor 3D structure [44,45]. The reader is referred to [73,74] for a comprehensive review.

QSAR
Quantitative structure-activity relationship (QSAR) is another popular strategy applied to the discovery of active molecules. QSAR consists of building mathematical models that statistically correlate the chemical structure with biological/toxicological properties by regression and classification methods [75,76]. In turn, this method can be categorized into 2D and 3D-QSAR. The 2D strategy calculates and compares compound properties in order to find similar molecules with which the compound is being compared (query molecule) [77]. On the other hand, 3D-QSAR relies not only on chemical structures but also on 3D coordinates of atoms to correlate with biological activity and is divided into alignment-based and alignment-independent techniques [78]. The reader is referred to [79][80][81] for a comprehensive review.

Development of PTP1B Inhibitors through Computational Approaches
In this section, we describe all the studies performed in the last century using a computational tool as the principal strategy to find PTP1B inhibitors and supported with in vitro studies.
In the year 2000, selective PTP1B inhibitors against leukocyte common antigenrelated phosphatase (LAR), receptor protein tyrosine phosphatase α (PTPα), and vaccinia H1-related protein phosphatase (VHR) were identified. These inhibitors were discovered by screening approximately 150,000 compounds with a virtual screening approach. From these compounds, twenty-five molecules were selected according to chemical diversity, interactions, overall fit with the enzyme active site, solubility, chemical stability, commercial availability, and cost. Seven compounds showed a high affinity for PTP1B (Ki = 21-510 µM); compound 2 (2-nitrobenzanthrone, Figure 3) was the most potent inhibitor (Ki = 21 µM) and exhibited selectivity against LAR, PTPα, and VHR (threefold higher for PTP1B). Furthermore, this compound was a mixed-type inhibitor [82]. Later, Doman and colleagues compared two methods to identify PTP1B inhibitors: virtual and high-throughput screening (VS and HTS, respectively). In this study, 235,000 and 400,000 compounds were assessed by VS and HTS, respectively. A total of 127 hits from VS and 85 hits from HTS showed IC 50 values <100 µM. Surprisingly, the most potent molecules were discovered by VS, and the hit rate was enhanced 1700-fold compared to HTS, with compound 1 (Figure 3) being the most notable molecule for its inhibitory activity (IC 50 = 4.1 µM) [48]. Two years later, another successful case was published by Lau and colleagues. They designed a series of benzotriazole phenyldifluoromethylphosphonic acids through molecular docking on the PTP1B crystal structure. Biphenylphosphonic acid (compound 19, Figure 3) was the most potent PTP1B inhibitor, with an IC 50 = 3 nM. In addition, this series of inhibitors was evaluated in several phosphatases, and a moderate selectivity against T-cell protein tyrosine phosphatase (TCPTP) was observed [83]. In 2005, PTP1B inhibitors were designed by optimizing the scaffold 1,2,5-thiadiazolidin-3-one-1,1-dioxide through molecular docking. The analysis indicated that a carbonyl moiety was necessary because it mimics the water-mediated interaction with PTP1B; additionally, the orthogonal orientation between the two rings was significant. These data were corroborated by NMR experiments and enabled identification of compound 10 (Figure 3), the most potent inhibitor (IC 50 = 2.47 µM) compared with the other designed molecules [84].
The following year, a series of monocyclic thiophenes were designed through the same computational strategy and guided by X-ray cocrystal structural information. It was observed that a hydrogen bond with Asp48 was key to achieving improved inhibition against PTP1B. Therefore, a carboxylic group was incorporated to promote an electrostatic interaction with Arg47, resulting in significant improvement in inhibitory activity (Ki = 0.14 µM, compound 36, Figure 4) and achieving a 236 and >1000 selectivity ratio against protein tyrosine phosphatase receptor type C (CD45) and LAR, respectively, although selective inhibition vs. TCPTP was not achieved (Ki = 0.18 µM) [85]. Two years later, these authors further pursued the optimization of thiophenes through molecular modeling, considering interactions with Arg24. The structural modification that they included was to remove the N-sulfonyl piperidine, and several side chains were employed to obtain hydrogen bonds with Arg24, which were confirmed by PTP1Binhibitor crystal structures. The analogous chloro derivative (compound 33, Figure 4) was the most potent inhibitor (Ki = 4 nM); however, its selectivity was not enhanced relative to TCPTP [86]. Wilson  chlorothiophene-2-carboxylic acid, Figure 4) was the most active, with subnanomolar activity against PTP1B (Ki = 0.00068 µM), and compound 32 (4-Bromo-3-carboxymethoxy-5-[3-(1phenylmethanesulfonylpiperidin-4-ylamino)phenyl]thiophene-2-carboxylic acid, Figure 4), with a Ki = 0.004 µM, resulted in significant selectivity against CD45 (Ki = 77 µM) and LAR (Ki = >500 µM) phosphatases. In addition, compound 54 (4-Bromo-3-carboxymethoxy-5-(3-{[1-(2,6-dimethylphenylcarbamoyl)piperidin-4-ylmethyl]amino}phenyl)thiophene-2carboxylic acid, Figure 4) showed selectivity between PTP1B and TCPTP (three times, Ki = 0.009 µM). Furthermore, pharmacokinetics studies determined that molecule 32 displayed an active glucose uptake mechanism into hepatocytes [87]. In the same year, pharmacophore and QSAR approaches were used to discover potent PTP1B inhibitors. Here, previously PTP1B inhibitors were employed to build a pharmacophoric model. The best binding hypothesis was integrated into a QSAR equation, and it was used to carry out a 3D search query to screen the National Cancer Institute database. Furthermore, the selected hits were filtered according to Lipinski's rules. The five compounds with the highest ranking were evaluated in vitro, with IC 50 values from nanomolar to low micromolar (0.47-3.30 µM), with compound 158 (Figure 4) being the most potent [88].  The following year, a series of monocyclic thiophenes were designed through the same computational strategy and guided by X-ray cocrystal structural information. It was observed that a hydrogen bond with Asp48 was key to achieving improved inhibition against PTP1B. Therefore, a carboxylic group was incorporated to promote an electrostatic interaction with Arg47, resulting in significant improvement in inhibitory activity (Ki = 0.14 µM, compound 36, Figure 4) and achieving a 236 and >1000 selectivity ratio against protein tyrosine phosphatase receptor type C (CD45) and LAR, respectively, although selective inhibition vs. TCPTP was not achieved (Ki = 0.18 µM) [85]. Two years later, these authors further pursued the optimization of thiophenes through molecular modeling, considering interactions with Arg24. The structural modification that they included was to remove the N-sulfonyl piperidine, and several side chains were employed to obtain hydrogen bonds with Arg24, which were confirmed by PTP1B-inhibitor crystal structures. The analogous chloro derivative (compound 33, Figure 4) was the most potent inhibitor (Ki = 4 nM); however, its selectivity was not enhanced relative to TCPTP [86]. Wilson  screen the National Cancer Institute database. Furthermore, the selected hits were filtered according to Lipinski's rules. The five compounds with the highest ranking were evaluated in vitro, with IC50 values from nanomolar to low micromolar (0.47-3.30 µM), with compound 158 (Figure 4) being the most potent [88]. In 2009, another study identified new PTP1B inhibitors through virtual screening. The docking library was taken from the latest version of the chemical database distributed by InterBioScreen. The compounds of this library were selected with drug-like filters and without reactive functional groups. This allowed for attainment of a database with 85,000 compounds instead of the 350,000 originals. Additionally, the authors modified the scoring function of the docking software, implementing a new solvation model. The 225 topscored compounds were evaluated; 21 compounds had more than 90% inhibition at 100 µM, 9 molecules were identified with IC50 values ranging from 10 to 50 µM, and compound 4 (thiazolidine-2,4-dione derivative, Figure 5) was found to be the most potent [49].  In 2009, another study identified new PTP1B inhibitors through virtual screening. The docking library was taken from the latest version of the chemical database distributed by InterBioScreen. The compounds of this library were selected with drug-like filters and without reactive functional groups. This allowed for attainment of a database with 85,000 compounds instead of the 350,000 originals. Additionally, the authors modified the scoring function of the docking software, implementing a new solvation model. The 225 top-scored compounds were evaluated; 21 compounds had more than 90% inhibition at 100 µM, 9 molecules were identified with IC 50 values ranging from 10 to 50 µM, and compound 4 (thiazolidine-2,4-dione derivative, Figure 5) was found to be the most potent [49]. On the other hand, Saxena Figure 5). The model predicted that the order of the inhibitory activity would be 3a > 3b > 3c, and the observed order was found to be similar: 3a (IC 50 = 69 µM) > 3c (IC 50 = 74 µM) > 3b (IC 50 = 87 µM). Furthermore, molecules 3a, 3b, and 3c administered at a dose of 100 mg/kg in two in vivo models decreased blood sugar levels by 25.1%, 19.8%, and 24.6%, in a sucrose-loaded rat model and 21.4%, 17.5%, and 20.6% in a streptozotocin-induced diabetic rat model, respectively [89]. A molecular- docking-guided design allowed for the synthesis of a series of di-indolinone derivatives. This strategy led to the discovery of PTP1B inhibitors with an IC 50 in the low micromolar range. Compounds 22 (1-[5-(5-bromo-2-Oxo-1,2-dihydroindol-3-ylidenemethyl)furan-2ylmethyl]-1H-indole-2,3-dione, Figure 5) and 32 (1-[5-(5-chloro-2-Oxo-1,2-dihydroindol-3-ylidenemethyl)furan-2-ylmethyl]-5-chloro-1H-indole-2,3-dione, Figure 5) showed high inhibitory activity (IC 50 = 2.8 µM and 2.3 µM, respectively). In addition, both molecules displayed selectivity over other homologous PTPs, such as TCPTP, Src homology-2 protein phosphatase-1 (SHP-1), and LAR of least 4 to 43 times, with IC 50 values for TCPTP of 11.5 and 18.8 µM, respectively, and >100 µM for the other phosphatases [90].
inhibitory activity would be 3a > 3b > 3c, and the observed order was found to be similar: 3a (IC50 = 69 µM) > 3c (IC50 = 74 µM) > 3b (IC50 = 87 µM). Furthermore, molecules 3a, 3b, and 3c administered at a dose of 100 mg/kg in two in vivo models decreased blood sugar levels by 25.1%, 19.8%, and 24.6%, in a sucrose-loaded rat model and 21.4%, 17.5%, and 20.6% in a streptozotocin-induced diabetic rat model, respectively [89]. A molecular-docking-guided design allowed for the synthesis of a series of di-indolinone derivatives. This strategy led to the discovery of PTP1B inhibitors with an IC50 in the low micromolar range. Compounds 22 (1-[5-(5-bromo-2-Oxo-1,2-dihydroindol-3-ylidenemethyl)furan-2-ylmethyl]-1H-indole-2,3-dione, Figure 5) and 32 (1-[5-(5-chloro-2-Oxo-1,2-dihydroindol-3-ylidenemethyl)furan-2-ylmethyl]-5-chloro-1H-indole-2,3-dione, Figure 5) showed high inhibitory activity (IC50 = 2.8 µM and 2.3 µM, respectively). In addition, both molecules displayed selectivity over other homologous PTPs, such as TCPTP, Src homology-2 protein phosphatase-1 (SHP-1), and LAR of least 4 to 43 times, with IC50 values for TCPTP of 11.5 and 18.8 µM, respectively, and >100 µM for the other phosphatases [90].  Figure 6) had the highest affinity for PTP1B (Ki = 6.7 µM), but it was not selective [91]. On the other hand, applying a high-throughput virtual screening strategy using the ZINC and IBS databases, a new inhibitor of PTP1B was discovered:  Figure 6) had the highest affinity for PTP1B (Ki = 6.7 µM), but it was not selective [91]. On the other hand, applying a high-throughput virtual screening strategy using the ZINC and IBS databases, a new inhibitor of PTP1B was discovered: ZINC ID: ZINC022765569 (3-(2-(1H-benzo[d]imidazol-2-ylthio)acetamido)benzoic acid, Figure 6). This compound inhibited 24% at 10 µM and decreased the glucose uptake by 18% in L6 muscle cells at 50 µM [92]. Furthermore, this inhibitor was taken as a starting point, and through molecular docking studies, a new series of PTP1B inhibitors was developed. Based on the predicted binding mode, different modifications were proposed, such as a methyl substitution at position 5 in the A ring, a substitution in the benzo ring of benzimidazole, and the replacement of the benzimidazole ring by phenyl oxadiazole. Two series of compounds were prepared, and the two most potent compounds, molecules 10c (3-(2-(5-Methoxy-1H-benzo[d]imidazol-2-ylthio)acetamido)-4-methylbenzoic acid, Figure 6) and 10e (2-Benzo[d]thiazol-2-ylthio)acetamido)-4-methylbenzoic acid, Figure 6), showed an IC 50 value of 8.2 and 8.5 µM, respectively [93]. Diphenyl ether derivatives were also identified as PTP1B inhibitors through virtual screening. The study was carried out in both PTP1B and TCPTP using an in-house compound database. From here, forty-three compounds were prioritized based on the docking scores. Compound AU-2439 (Figure 6) was the most potent, with an IC 50 of 43 µM. Additionally, it was found to be a selective inhibitor with fivefold selectivity for PTP1B over TCPTP (IC 50 =230 µM) [94]. In another study, PTP1B inhibitors were discovered via an integrated molecular design strategy of pharmacophore-oriented scaffold hopping based on the template structure of Ertiprotafib. The information obtained from the interaction mode of Ertiprotafib with PTP1B simulated by molecular docking helped to develop a pharmacophore model for Ertiprotafib. Accordingly, twenty-one molecules from five distinct structural classes were designed and synthesized. Of these, nine molecules significantly inhibited PTP1B with a percentage of inhibition higher than 80% at 100 µM. The two most active compounds were 3a and 4e ( Figure 6), exhibiting an IC 50 value of 1.3 and 3.9 µM, respectively [46]. Imidazolidine-2,4dione derivatives were reported as selective PTP1B inhibitors using virtual screening and molecular docking with the core hopping method. The core hopping algorithm helped to find the cores attaching to the scaffold using fragments from the ZINC database. A total of fifty molecules were docked at the active site of PTP1B and TCPTP and; consequently, twelve compounds were synthesized and assessed in both PTP1B and TCPTP. Overall, most molecules were potent and selective, with the compound designed as #h (Figure 6) being the most selective, with a selectivity index of almost 34 times PTP1B (IC 50 =4.1 µM) over TCPTP (IC 50 > 130 µM) [95].
In 2020, Wu and colleagues identified a series of novel PTP1B inhibitors using a virtual screening workflow, molecular docking, and ADMET (absorption, distribution, metabolism, excretion, and toxicity) strategy. Compound ZINC39276654 was selected from the ZINC database by virtual screening approach to generate one thousand molecule derivatives. Subsequently, one hundred generated derivatives were selected among the top-scored molecules and redocked into the PTP1B active site to choose the top ten compounds (1a-1j) to evaluate the inhibitory capacity. Compound 1a (ethyl 2-(2-bromophenyl)-4-((4methoxyphenoxy)methyl)thiazole-5-carboxylate, Figure 10) exhibited the best inhibitory potency (IC 50 = 4.46 µM) and acceptable predicted pharmaceutical properties. This molecule was selective for PTP1B over other phosphatases-3 times for TCPTP (IC 50 = 12.28 µM) and >22 times for the rest-Src homology region 2 domain-containing phosphatase 1 (SHP1), megakaryocyte protein tyrosine phosphatase 2), CDC25B (M-phase-inducer phosphatase 2 (MEG2), and LAR. However, compound 1g (ethyl 4-((4-chlorophenoxy)methyl)-2-(3phenoxyphenyl)thiazole-5-carboxylate, Figure 10) showed the strongest selectivity for PTP1B over the TCPTP (>9 times, IC 50 = >100 µM) [106]. In the same year, a series of seventeen novel 4-thiazolinone derivatives that shared the scaffold of compound ZINC99459 was discovered by virtual screening. ADMET prediction (including human intestinal absorption, blood-brain barrier, plasma protein binding, aqueous solubility, cytochrome P450 2D6 binding, and toxicity) indicated that these compounds showed good drug-likeness properties. Then, in vitro enzyme inhibitory activity was determined on PTP1B and other phosphatases. Compound 7p (Figure 10) was identified as the most potent (IC 50 = 0.92 µM) and selective inhibitor, >100 times (IC 50 = >100 µM) against TCPTP, SHP2, CDC25B, LAR, and MEG2 and at least 24 times against SHP1 [107]. Another effort to find PTP1B inhibitors via computational tools was a work reported by Yang and colleagues. A natural product library of approximately 122,000 compounds was screened by different in silico filters, such as the 3D-QSAR model, 2D fingerprint similarity, and molecular docking. The twenty-six molecules prioritized by virtual screening were explored to determine their inhibition against PTP1B. The most active compounds, coumarin derivatives, showed an IC 50 in the micromolar range. Among them, nine molecules were further evaluated against several phosphatases showing selective inhibition for PTP1B over the other phosphatases, including TCPTP. The most active inhibitor was H17 (IC 50 = 2.05 µM, Figure 10), but it was not selective (inhibitory activity on TCPTP: 73.5% at 10 µM). The most selective compounds were H8 and H20 (Figure 10), with a selectivity of approximately 10 times (inhibitory activity < 10% at 10 µM) with respect to TCPTP, CD45, LAR, and VHR phosphatases [108]. In 2020, Wu and colleagues identified a series of novel PTP1B inhibitors using a virtual screening workflow, molecular docking, and ADMET (absorption, distribution, metabolism, excretion, and toxicity) strategy. Compound ZINC39276654 was selected from the ZINC database by virtual screening approach to generate one thousand molecule derivatives. Subsequently, one hundred generated derivatives were selected among the topscored molecules and redocked into the PTP1B active site to choose the top ten compounds (1a-1j) to evaluate the inhibitory capacity. Compound 1a (ethyl 2-(2-bromophenyl)-4-((4-methoxyphenoxy)methyl)thiazole-5-carboxylate, Figure 10) exhibited the best inhibitory potency (IC50 =4.46 µM) and acceptable predicted pharmaceutical properties. This molecule was selective for PTP1B over other phosphatases-3 times for TCPTP (IC50 = 12.28 µM) and >22 times for the rest-Src homology region 2 domain-containing In 2021, the scaffold-hopping approach was employed to optimize the structure of compound 1 based on the replacement of the pyrrole ring by azoles. The new compounds showed better drug-like properties than compound 1 according to the results obtained by means of in silico tools. Most molecules showed good inhibitory activity within the IC50 range of 0.46-2.17 µM. Among these compounds, compound 2 (4-methylimidazo[1,2a]quinoxaline, IC50 = 0.49, Figure 11) and compound 9 (tetrazolo[1,5-a]quinoxaline, IC50 = 0.62, Figure 11) displayed the highest potency. Additionally, selectivity evaluation of these compounds against TCPTP showed that the most selective was compound 2 (two times, inhibitory activity on TCPTP: 44% at 1 µM). This molecule also increased glucose uptake by 15% relative to control cells in phenotypic models [109]. Ma and colleagues discovered imidazolidine-2,4-dione derivatives as PTP1B inhibitors with structural diver- Figure 10. Structure of the most potent PTP1B inhibitors reported in 2020 discovered through computational approaches.
In 2021, the scaffold-hopping approach was employed to optimize the structure of compound 1 based on the replacement of the pyrrole ring by azoles. The new compounds showed better drug-like properties than compound 1 according to the results obtained by means of in silico tools. Most molecules showed good inhibitory activity within the IC 50 range of 0.46-2.17 µM. Among these compounds, compound 2 (4methylimidazo[1,2-a]quinoxaline, IC 50 = 0.49, Figure 11) and compound 9 (tetrazolo[1,5a]quinoxaline, IC 50 = 0.62, Figure 11) displayed the highest potency. Additionally, selectivity evaluation of these compounds against TCPTP showed that the most selective was compound 2 (two times, inhibitory activity on TCPTP: 44% at 1 µM). This molecule also increased glucose uptake by 15% relative to control cells in phenotypic models [109]. Ma and colleagues discovered imidazolidine-2,4-dione derivatives as PTP1B inhibitors with structural diversity using virtual screening, scaffold hopping, ADMET prediction, and molecular docking. First, the compound CHEMBL213560, obtained by virtual screening, was optimized using the scaffold-hopping method. The designed compounds, with good ADMET results, were used to develop a molecular docking anal-ysis. Fifteen compounds with high scores for PTP1B protein were selected to evaluate their biological activity. These molecules were assessed against PTP1B, Src homology region 2 domain-containing phosphatase 2 (SHP2), and LAR; compound 10 ((E)-4-(methoxycarbonyl)benzyl 4-((3-benzyl-4-(4-((4-(methoxycarbonyl)benzyl)oxy)benzylidene)-2,5-dioxoimidazolidin-1-yl)methyl)benzoate, Figure 11) the highest PTP1B inhibition, with an IC 50 of 2.07 µM. Moreover, this molecule had 10 times higher inhibitory capacity on PTP1B than SHP2 and 60 times higher than that of LAR [110]. Finally, another study of integrated virtual screening consisting of fingerprint similarity search, structure-based pharmacophore models, and molecular docking was carried out to search for potential allosteric PTP1B inhibitors from commercially available chemical libraries. Of 184,922 molecules, nine compounds were selected to be evaluated in vitro. Two compounds were the most active: H3 (IC 50 = 0.72 µM, Figure 11) and H9 (IC 50 = 1.59 µM, Figure 11). They were further evaluated against TCPTP and SHP2, displaying selective inhibition of PTP1B over both phosphatases (>60 times, IC 50 = >100 µM) [111].

Conclusions
A considerable number of PTP1B inhibitors has been reported as outstanding and potential hits or leads for new drugs against type 2 diabetes. Several potent and selective PTP1B inhibitors have been identified via different computational tools. These studies reported compounds with IC50 and Ki values from the nanomolar to the micromolar range.

Conclusions
A considerable number of PTP1B inhibitors has been reported as outstanding and potential hits or leads for new drugs against type 2 diabetes. Several potent and selective PTP1B inhibitors have been identified via different computational tools. These studies reported compounds with IC 50 and Ki values from the nanomolar to the micromolar range. Furthermore, some of these compounds were found to be selective on PTP1B over TCPTP, its closest homolog, with a selectivity range of 2 to 100 times. Moreover, when the characterization was far away, they showed the ability to improve glucose uptake. The PTP1B inhibitor design process remains challenging, which has led to many academic research laboratories and pharmaceuticals organizations continuing the search for effective and safe oral available compounds. Therefore, this review encourages the use of computational techniques and includes helpful information that increases the knowledge generated to date about PTP1B inhibition, with a positive impact on the route toward obtaining a new drug against type 2 diabetes with PTP1B as a molecular target.