Discovery of Non-Inhibitory Macrocyclic Ligands for Protein Tyrosine Phosphatase 1B Using a Function-Based, Iterative Screening Strategy

Abstract

Chemically induced proximity is a powerful modality for manipulating protein function. Most of the effort in this field has focused on targeted protein degradation but recruitment of other types of post-translational modification enzymes to a target protein is also of interest. To construct such reagents, one would ideally like to have ligands that engage the enzyme without inhibiting its activity. In this study, we describe a screening platform for the discovery of non-inhibitory macrocyclic ligands for a protein tyrosine phosphatase, using PTP1B as an exemplary model target. This workflow involves sequential screens of small libraries of bead-displayed macrocycles in which only one position of the macrocycle is varied in each round of screening while the others are held as invariant placeholders. The beads co-display a high K_M substrate for the phosphatase, allowing ligand-dependent recruitment of the enzyme to the bead surface to be coupled to dephosphorylation of the co-displayed substrate. This is detected by staining with a labeled anti-phosphotyrosine antibody. Finally, we demonstrate that the same general approach can be applied to proteins lacking enzymatic activity by screening against biotin ligase-target protein fusions and employing a proximity labeling-like assay to register screening hits.

Graphical Abstract

INTRODUCTION

Proteolysis-targeting chimeras (PROTACs)¹ and molecular glue degraders² are of considerable current interest as an alternative to traditional, occupancy-driven inhibitors. These compounds recruit a target protein and an E3 Ubiquitin ligase complex into a ternary complex that results in poly-Ubiquitylation and subsequent proteasome-mediated destruction of the target. There are many other post-translational modifications (PTMs) besides poly-Ubiquitylation that would be of interest to trigger using proximity-inducing molecules.³ In particular, protein phosphorylation and dephosphorylation are key events in regulating the function of many proteins. In particular, cascades of phosphorylation events are critical to virtually every signal transduction pathway employed by cells to couple extracellular signals to changes in gene expression. Thus, the development of PHICs (phosphorylation-inducing chimeras)⁴ and PHORCs (phosphatase-recruiting chimeras)⁵ capable of recruiting a kinase or phosphatase to a given target protein may constitute a novel approach to regulating flow through these pathways.

For the construction of such reagents, one would ideally like to employ ligands that engage the kinase or phosphatase in a manner that does not inhibit its catalytic activity. Unfortunately, the vast majority of existing compounds that engage these enzymes are inhibitors, having been discovered in high-throughput screening campaigns using functional assays. “Innocent” ligands cannot be distinguished from inactive compounds with such methods. Some ligand discovery technologies, such as DNA-encoded libraries (DELs),⁶ mRNA⁷ and phage display,⁸ or fragment-based drug discovery (FBDD),⁹ are better suited for this purpose since they simply register binding of a ligand to a target protein, but inhibitors will also score as hits in screens using these techniques. Thus, there is a need for the development of new high-throughput screening platforms that are strongly biased towards identification of innocent ligands and enzyme activators.

We recently reported a function-based one-bead one-compound (OBOC) screening platform for identifying non-inhibitory E3 ligase ligands, in which library beads co-display a candidate ligand and a neo-substrate.¹⁰ Ligand-dependent recruitment of an E3 ligase complex drives proximity-mediated neo-substrate poly-Ubiquitylation. By coupling the binding event to modification of the co-displayed substrate, this platform is biased towards the recovery of innocent ligands or activators and against inhibitors. We hypothesized that this platform could be extended to the discovery of non-inhibitory ligands for other types of PTM enzymes by simply changing the nature of the co-displayed substrate and the labeled antibody employed to detect the modified substrate (Scheme 1). In this study we show that this is indeed the case for the discovery of ligands to a protein tyrosine phosphatase (PTP). PTP1B¹¹ is an abundant and ubiquitously expressed PTP, and so was chosen as a suitable initial target protein for this application.

Scheme 1.

General format for the discovery of bead-displayed, non-inhibitory enzyme ligands. Ligand-dependent recruitment of an enzyme to the bead surface results in proximity-driven modification of a peptide or protein substrate co-immobilized on the bead surface. The resulting PTM provides a covalent mark that can be detected using a fluorescently labeled anti-PTM antibody (red) after a rigorous wash. The green balls represent pieces of a macrocyclic ligand that engages the enzyme outside the active site and does not interfere with its activity.

A potential complication with this approach is that the active site is often the only well-defined binding pocket on PTM enzymes, including PTPs, making it challenging to engage alternative, non-catalytic surfaces. In this context, beyond-Rule-of-Five (bRo5) macrocycles,¹² with their larger molecular “wingspan” and expanded interaction surface, represent an attractive chemical space for identifying ligands that can engage shallow pockets outside the active site without inhibiting enzyme activity. Therefore, in this study, we focused our efforts on the discovery of PTP1B-binding, non-inhibitory macrocycles.

Finally, the development of certain types of chemical dimerizers will require recruitment of proteins that lack enzymatic activity or whose activity is not readily monitored in this format. To allow the extension of this assay platform (Scheme 1) to such targets, we show here that one can employ as the screening target a fusion of the protein of interest to a biotin ligase (TurboID¹³) and monitor proximity-induced biotinylation of a co-displayed, lysine-containing peptide.

RESULTS

Establishment of a functional assay for the discovery of non-inhibitory PTP ligands

10 μm TentaGel beads were modified such that ≈ 15% of the sites displayed a phosphopeptide substrate for PTP1B and the remaining sites displayed an ethylene glycol-containing linker capped with a potential ligand (Figure 1; see Methods for synthetic details). Specifically, we examined two previously reported PTP1B substrates: the EGFR-derived phosphopeptide DADEpYLIPQQG and an artificial peptide substrate, DHVTQpYAA which have reported K_M values of 4 μM and >200 μM, respectively.¹⁴ To assess the extent of ligand-independent dephosphorylation of the bead-displayed substrates, we prepared control beads displaying only an acetyl group, referred to as beads 1P and 2P, where “P” indicates the co-displayed peptide contains a phosphotyrosine. To benchmark the maximum extent of on bead dephosphorylation achievable, we also synthesized control beads displaying the corresponding non-phosphorylated substrate peptides together with an acetyl group (Beads 1 and 2).

Figure 1.

Design of assay beads for on bead dephosphorylation characterization and validation.

Beads 1P and 2P were incubated with different concentrations of recombinant GST-PTP1B for two hours, after which time the beads were washed thoroughly, then stained with AlexaFluor 647 (A647)-labeled anti-phosphotyrosine antibody. After another wash, the beads were analyzed by flow cytometry. As shown in Figures 2A and 2B, the fluorescent signal in the A647 channel of the beads was lower than that of the analogous resin not treated with the phosphatase, indicating some level of ligand-independent dephosphorylation. Quantitatively, however, the high and low K_M peptide substrates behaved quite differently. Beads 1P displaying the EGFR-derived substrate exhibited a significant decrease in anti-phosphotyrosine antibody staining upon incubation with as little as 4 nM PTP1B. Nearly complete dephosphorylation (represented by the signal of beads 1) was observed after incubating with 20 nM PTP1B. In contrast, beads 2P, displaying the high K_M artificial substrate, exhibited only limited loss of tyrosine phosphorylation even after incubation with up to 500 nM PTP1B. The fluorescent signal remained more than three orders of magnitude higher than that measured for the beads displaying the corresponding non-phosphorylated substrate. These results make clear that a high K_M substrate must be employed to provide a sufficiently large window to distinguish beads that display peptides that have been dephosphorylated from those that have not.

Figure 2.

Establishment and validation of a proximity-driven labeling screening platform using GST-PTP1B and GSH as a model target-ligand pair. (A) Flow cytometry analysis of Beads 1P after incubation with GST-PTP1B at the indicated concentrations, followed by staining with Alexa Fluor 647-conjugated anti-phosphotyrosine antibody. The corresponding non-phosphorylated control beads (Beads 1) were also analyzed to define the signal associated with complete dephosphorylation. (B) Flow cytometry analysis of Beads 2P-4P after incubation with GST-PTP1B at the indicated concentrations, followed by staining with Alexa Fluor 647-conjugated anti-phosphotyrosine antibody. The corresponding non-phosphorylated control beads (Beads 2-4) were also analyzed to define the signal associated with complete dephosphorylation. (C) Flow cytometry analysis of Beads 2P-4P after incubation with His-PTP1B at the indicated concentrations, followed by staining with Alexa Fluor 647-conjugated anti-phosphotyrosine antibody. (D) Flow cytometry analysis of a 1:1:1 mixture of Beads 2P, 3P, and 4P before and after incubation with GST-PTP1B at the indicated concentrations, followed by staining with Alexa Fluor 647-conjugated anti-phosphotyrosine antibody.

The other side of the coin is to measure the level of ligand-dependent de-phosphorylation of the high K_M substrate using beads displaying a positive control and determine the degree of correlation between ligand-protein binding affinity and the degree of on-resin dephosphorylation of the co-displayed substrate. We are not aware of any non-inhibitory PTP1B ligands that could be used for this purpose. Therefore, we employed GST ligands as surrogates for such compounds. The co-crystal structure of the GST-glutathione (GSH) complex (PDB: 1UA5) indicates that modification of the thiol group of the cysteine residue in GSH is well tolerated.¹⁵ In contrast, the free carboxylic acid of the glycine residue participates in ionic interactions with GST, and modification at this position is expected to reduce binding affinity significantly. To test this hypothesis, we synthesized GSH derivatives bearing FITC labels to either the cysteine thiol or the glycine carboxylate (Compound 1 and 2, respectively; structures shown in Figure S1A) and evaluated their binding to GST-PTP1B by titration followed by a change in fluorescence polarization (FP). Compounds 1 and 2 were found to bind GST-PTP1B with K_D values of 0.5 μM and 7 μM, respectively (Figure S1B). Thus, attachment of GSH to the resin via either the cysteine thiol or the C-terminal carboxyl unit will provide models of high and low affinity, non-inhibitory, bead-displayed ligands for GST-PTP1B.

Beads 3P and 4P, as well as their non-phosphorylated analogues (beads 3 and 4; see Figure 1) were synthesized and incubated with GST-PTP1B at concentrations ranging from 125 nM to 500 nM for two hours, after which the degree of dephosphorylation of the co-displayed, high K_M substrate was analyzed by antibody staining and flow cytometry. Gratifyingly, in all cases we observed readily measurable differences in fluorescence between beads 3P and 4P relative to the acetylated beads 2P (Figure 2B). Beads 4P, displaying the high affinity GST ligand tethered to the bead via a thioether linkage, were almost completely dephosphorylated at all GST-PTP1B concentrations examined. Beads 3P, displaying the lower affinity ligand showed an enzyme concentration-dependent level of dephosphorylation. Compared to acetylated negative control beads 2P, incubation of the beads with His-PTP1B (lacking the GST fusion) did not result in any detectable enhanced dephosphorylation on ligand displaying beads 3P and 4P (Figure 2C), as expected.

Finally, to validate that the different signal intensities observed in the above model experiment would be suitable to support a ligand discovery effort, we conducted a mock screen in which beads 2P, 3P and 4P were mixed in approximately equal ratios and incubated with GST-PTP1B (125 nM or 500 nM), followed by the same washing and analysis steps described above. To better analyze the separation of bead populations, the data were plotted as contour plots in which the y-axis represents AlexaFluor 647 fluorescence intensity (corresponding to bead phosphorylation levels) and the x-axis shows the total FSC signal, which facilitates visualization. As shown in Figure 2D, before enzyme incubation, the mixed beads shared comparable phosphorylation levels and formed a single population on the contour plot. After incubation with 125 nM GST-PTP1B, the mixed beads formed three clearly separated populations, and the population distribution aligned well with the approximately 1:1:1 ratio of the three bead types. At 500 nM GST-PTP1B, the difference in fluorescence intensity between the populations we presume are 3P and 4P narrowed and there was some modest drop in intensity in the fluorescence of the third population, presumed to be beads 2P.

We conclude from this model experiment that this assay format displays an excellent signal to noise ratio and is capable of detecting both high (500 nM) and relatively low (7 μM) affinity ligands.

Discovery of a Macrocyclic PTP1B Ligand

With a suitable screening assay established, we turned to the discovery of novel, non-inhibitory PTP1B ligands from a library of bead-displayed macrocycles. In this initial study, we opted to aim for the identification of a macrocyclic peptide (MP) ligand for PTP1B because of the low cost and ready availability of many different amino acid building blocks. Because most MPs exhibit poor cell permeability, we incorporated an imidazopyridinium (IP⁺) heterocycle into each ring. The IP⁺ unit reliably improves the passive cell permeability of MPs, often dramatically.^{16, 17} To avoid the necessity of having to screen very large libraries of macrocycles, we explored the workflow shown schematically in Figure 3. In the first round of screening, only one site (“position 1”; shown in green) in the macrocycle is varied while all of the others are held as invariant placeholders. In a second round of screening, the best hit from round one is held invariant at position 1 (red), while position 2 (green) is varied. Again, the remaining positions are held invariant as placeholders. This protocol is then repeated until a ligand for PTP1B is in hand. This workflow is somewhat similar to that employed in fragment-based drug development (FBDD),^{9, 18} where the initial screen is directed towards the discovery of a low affinity ligand that is then “grown” into a higher affinity lead compound,¹⁹ though these macrocycles are obviously not fragments. A potential complication with this approach is that the round one hits will almost certainly be weak ligands for PTP1B, with affinities likely to be in the high μM to low mM range. Indeed, FBDD campaigns require using highly sensitive binding assays, such as NMR or SPR to identify protein-binding fragments. We hoped that the assay described above would be able to register hits with weak binding affinities since the enzyme presumably dephosphorylates several co-displayed substrates when attracted to the bead, thus providing an amplification of each binding event. The fact that the modest affinity GST ligand supported a level of dephosphorylation readily distinguishable from the negative control at least partially supports this hypothesis, though its 7 μM K_D is probably much lower than that of any initial screening hits we would expect to obtain.

Figure 3.

Schematic representation of the multi-step, sequential screening workflow for the discovery of a macrocycle ligand. Black balls represent invariant placeholders, green balls a position that is varied and red balls an invariant position that is occupied by the best hit from the previous round of screening.

The bead architecture and synthetic workflow for the first-generation library are shown in Figure 4. Specifically, peptide H₂N-Gly-Gly-Gly-Lys(Mmt) was first synthesized on 10 μm TentaGel beads bearing approximately 15% azide-terminated PEG handles, with the Mmt-protected lysine directly coupled to the remaining amine handles on the beads. The beads were then split into a 96-well microtiter filter plate, with approximately 250,000 beads per well. In 87 wells, a distinct amino acid building block was coupled to the N-terminus of the resin-bound Gly-Gly-Gly-Lys(Mmt) peptide in each well (see Figure S2 for the amino acids used). In eight of the remaining wells, glycine was coupled in the same manner to serve as a negative control. After this diversification step, a serine residue was added to the bead-bound peptides in all wells, which could be oxidized by NaIO₄ to provide N-terminal oxo-aldehyde. Following oxidation, the Mmt protecting group at lysine side chain was deprotected to form a cyclic imine intermediate. 2-formylquinoline was then used to close the ring through formation of the IP⁺ heterocycle.¹⁶ Finally, the azide handles on the beads in each well were functionalized in parallel with a fluorescein-labeled, low-affinity artificial PTP1B substrate peptide via copper-catalyzed azide-alkyne cycloaddition (CuAAC).²⁰. Beads in four of the eight glycine-control wells were instead functionalized with the corresponding non-phosphorylated substrate to benchmark complete on-bead dephosphorylation.

Figure 4.

Design and synthesis of one-bead-one-compound (OBOC) libraries displaying IP⁺-containing macrocyclic peptides (MPs). (A) Schematic illustration of the assay bead design. (B) General workflow for the synthesis of the proposed first generation OBOC libraries.

To screen this library, the beads in each well were individually incubated with 200 nM His₆-PTP1B for 2 hours. The reaction was then terminated by washing away enzyme and the beads were stained with an Alexa Fluor 647 conjugated anti-phosphotyrosine antibody. After another wash, the remaining phosphotyrosine levels on the beads in each well were quantified by measuring Alexa Fluor 647 fluorescence using a plate reader. The level of fluorescein intensity, which represents the total amount of substrate peptide loaded, was also measured, allowing a normalization check to be made, correcting for possible variation in substrate levels on the beads in different wells.

The results are shown in Figure 5A. Compared to the negative control macrocycle containing glycine at all positions, a total of 9 bead populations exhibited a level of fluorescence suggesting at least 15% dephosphorylation of the co-displayed peptide substrate. The macrocycle with (R)-3-phenyl-β-alanine (Building Block A10) at the variable position (henceforth called compound 3) displayed the highest level (50%) of dephosphorylation of the co-displayed peptide. A fluorescent derivative of 3 with a FITC label (3-FITC) was synthesized and titrated with PTP1B. Binding was detected by an increase in fluorescence polarization (FP). This resulted in a dose-dependent increase in the FP signal for compound 3-FITC (Figure 5E), confirming interaction with PTP1B. However, saturation was not achieved even at 150 μM PTP1B, indicating that 3 is a weak ligand with a K_D > 100 μM, as expected. This experiment demonstrates that the on-resin dephosphorylation assay is sufficiently sensitive to detect even very low affinity ligands. Interestingly, several bead populations showed residual phosphotyrosine levels significantly higher than those observed for the four-glycine macrocycle negative control. Since the four-glycine IP⁺ macrocycle itself exhibits weak, but non-negligible, binding to PTP1B (Figure S4), we speculate that certain building blocks may disrupt this basal interaction, thereby reducing PTP1B recruitment and resulting in higher residual phosphotyrosine signals.

Figure 5.

Screening of the first two generations of MP-displaying OBOC libraries and characterization of the optimal hits. (A) Scatter plot summarizing the normalized residual phosphotyrosine levels on beads displaying individual compounds from the first-generation OBOC library, in which variable position 1 was diversified, after incubation with 200 nM His-PTP1B. Signals were normalized to the phosphotyrosine signal from beads displaying the four-glycine macrocycle negative control. The horizontal dashed line indicates 15% dephosphorylation. The optimal hit is highlighted with a red box. (B) Structures of the optimal hit from the first-generation library (Compound 3) and its FITC-labeled FP probe (Compound 3-FITC). (C) Scatter plot summarizing the normalized residual phosphotyrosine levels on beads displaying individual compounds from the second-generation OBOC library, in which variable position 2 was diversified, after incubation with 200 nM His-PTP1B. Signals were normalized to the phosphotyrosine signal from beads displaying the four-glycine macrocycle negative control. The horizontal dash line represents the averaged remaining phosphorylation level of beads displaying the first-generation hit 3. The optimal hit is highlighted with a red box, Additional hits selected for evaluation are highlighted with black boxes. (D) Structures of the optimal hit from the second-generation library (Compound 4) and three FP probes derived from Compound 4 (Compound 4-FITC, Compound 5, and Compound 6). (E) FP titration curves of Compound 3-FITC and Compound 4-FITC with His-PTP1B. (F) FP titration curves of Compound 4-FITC, Compound 5, and Compound 6 with His-PTP1B.

A second-generation library was then synthesized in which (R)-3-phenyl-β-alanine was set as an invariant residue at position 1 and the adjacent position was diversified using 87 different amino acids (Figure S2). The remaining two positions in the ring were again invariant glycine placeholders. In addition to the previously described four-glycine macrocycle displaying control beads, beads displaying the first-generation hit (compound 3) were included in the plate as an additional benchmark to facilitate the identification of improved derivatives. This second-generation library was synthesized in duplicate and screened using the same workflow as the first-generation library, but at PTP1B concentrations of 200 nM and 50 nM. The additional “low PTP1B” screening condition was introduced to improve resolution for ligands with significantly higher PTP1B binding affinity. In this round, only beads exhibiting lower phosphorylation levels than those displaying compound 3 at both enzyme concentrations were defined as hits and advanced for further analysis.

As shown in Figure 5C, after incubation with 200 nM PTP1B, 32 bead populations exhibited greater dephosphorylation than beads displaying the first-generation hit 3, which is indicated by the horizontal line. Among these, the macrocycle containing 3,4-dichlorophenylalanine (Building Block B1) at the second variable position, hereafter referred to as compound 4, produced the highest level of dephosphorylation and showed 30% residual phosphorylation relative to beads displaying the four-glycine macrocycle. This trend was also observed in an analogous screen performed at a lower PTP1B concentration of 50 nM (Figure S3), further supporting compound 4 as the top hit from this round of screening. To validate that this increased dephosphorylation corresponds to a higher affinity for PTP1B, a FITC-labeled derivative of compound 4 was synthesized and titrated with the enzyme. As shown in Figure 5E, a dose-dependent increase in FP signal was observed, yielding a K_D of 14 μM for the PTP1B·4-FITC complex. This represents a substantial improvement in binding affinity relative to the first-generation hit, compound 3.

In addition to Building Block B1, several other building blocks, including A2 (3-(2-pyridyl)-L-alanine) and E1 (4-amino-3(s)-hydroxybutanoic acid), also yielded compounds in the second-generation library that appeared to recruit PTP1B with efficiencies comparable to that of compound 4-displaying beads in the screen employing 200 nM PTP1B. However, at an enzyme concentration of 50 nM, these bead populations retained substantially higher phosphorylation levels than compound 4-displaying beads, suggesting that the corresponding macrocyclic peptides bind PTP1B more strongly than compound 3 but more weakly than compound 4. To confirm this interpretation, we synthesized FITC-labeled compounds 23 and 24 (structures shown in Figure S4A), which contain Building Blocks A2 and E1 at position 2, respectively, and evaluated their PTP1B binding affinities. As expected, the K_D values of the PTP1B•23 and PTP1B•24 complexes fell between those of compounds 3 and 4, with values of 74 and 69 μM, respectively (Figure S4B). To further demonstrate that the assay readout correlates with ligand affinity, we also synthesized a FITC-labeled derivative of compound 25 (structures shown in Figure S4A), which contains building block E11 (D-β-homoproline) at the second variable position. Beads displaying this compound showed 74% residual phosphorylation relative to beads displaying the four-glycine macrocycle in the second-round screen with 200 nM PTP1B. As expected, the affinity of 25 for PTP1B fell between that of compound 3 and the four-glycine macrocycle (Figure S4B), further supporting the ability of the on-bead dephosphorylation assay to report relative ligand affinity.

We then moved on to create a third-generation library in which (R)-3-phenyl-β-alanine was placed at position 1, 3,4-dichlorophenylalanine at position 2, and position 3 on the macrocycle was varied. Position 4 was again held constant using a glycine placeholder. In this case, no compounds with significantly improved ability to recruit PTP1B were identified (Figure S5). This result suggested that position 3, and perhaps position 4, were solvent exposed and not oriented to allow side chains at these positions to make productive contacts with the phosphatase. To probe this hypothesis further, a bulky FITC-labeled lysine residue was introduced at position 3 or position 4 (compounds 5 and 6, respectively), and the affinity of these molecules for PTP1B was measured, reasoning that steric interference would weaken binding if these units were oriented towards the enzyme but have little or no effect if they are indeed solvent exposed. As shown in Figure 5F, 5 and 6 engaged PTP1B with K_D values of 10 μM and 8 μM, comparable to the 14 μM K_D of the FITC-labeled second-generation hit compound 4. We conclude that positions 3 and 4 are indeed solvent exposed.

While positions 3 and 4 in the ring are apparently not poised to make productive contacts with PTP1B, the IP⁺ unit may be since it is adjacent to position 1 in the macrocycle. Therefore, 11 different 2-formyl pyridine derivatives were employed in the macrocyclization reaction¹⁶ to create FITC labeled macrocycles 7-17 (7-FITC to 17-FITC), as illustrated at the top of Table 1. In this case, given the small number of compounds, we proceeded directly to a titration experiment followed by an increase in FP to determine their affinities for PTP1B rather than carry them through the dephosphorylation assay. The best compound, macrocycle 7, in which an indole ring is fused to the IP⁺ heterocycle, bound PTP1B with a K_D of 1 μM (Table 1 and Figure S6), a 14-fold improvement over the affinity of macrocycle 4 for the phosphatase.

Table 1.

Structures of FITC-labeled Compound 4 derivatives and their binding affinities for His-PTP1B.

Macrocycle 7 is a selective, cell permeable, non-inhibitory ligand for PTP1B

Since macrocycle 7 was identified using screens that required PTP1B to dephosphorylate a substrate when recruited to the bead, we presumed that it is either an activator or an innocent ligand, not an inhibitor. However, recent reports of modest potency inhibitors functioning as useful ligands in the construction of functional kinase- and deubiquitylase-recruiting chimeras²¹ call this assumption into question. In these studies, it was hypothesized that dissociation of the enzyme from the inhibitor that recruited it allows some degree of localized activity. Therefore, we first assessed the effect of 7 on PTP1B activity in vitro. In this assay,²² PTP1B-catalyzed hydrolysis of the fluorogenic substrate DiFMUP was monitored in the presence of serially diluted compound 7, allowing any inhibitory activity to be quantified from the reduction in fluorescence signal relative to the no-inhibitor control. As shown in Figure 6A, macrocycle 7 did not inhibit PTP1B-catalyzed dephosphorylation activity even at 50 μM, a concentration 50-fold greater than the K_D of the complex, whereas the known PTP1B inhibitor, PTP1B-IN-4, inhibited the enzyme with an IC₅₀ of 12 μM, consistent with previously reported values.

Figure 6.

Characterization of the optimal screening hit Compound 4 and its derivative Compound 7. (A) IC₅₀ curve of Compound 7 and known PTP1B inhibitor PTP1B-IN-4. (B) FP titration curves of Compound 4-FITC with PTP1B, TC-PTP, and SHP2. (C) FP titration curves of Compound 7-FITC with PTP1B, TC-PTP, and SHP2. (D) Structures of chloroalkane tagged compounds tested in CAPA assay. (E) CAPA data for Compound 4-Ct and Compound 7-Ct as well as corresponding control MPs lacking the IP⁺ unit (Compound 18 and 19). Chloroalkane-tagged JQ1 (JQ1-Ct) was included as a permeability benchmark. Measurements were performed in triplicate.

The development of orthosteric phosphatase inhibitors is notoriously difficult due to the high degree of sequence conservation in and around the active site²³ However, since macrocycle 7 is not an inhibitor, it must engage PTP1B outside of the active site, raising the possibility that it might exhibit significant selectivity. To probe this, we measured the binding affinity of 7-FITC, as well as 4-FITC, for T cell protein tyrosine phosphatase (TC-PTP) (residues 1–381), the human phosphatase most closely related to PTP1B.²⁴ The catalytic domains of these enzymes are 74% identical. As shown in Figure 6B and 6C, both macrocycles exhibited selective binding to PTP1B. 7-FITC bound PTP1B about 23 times more tightly than TC-PTP. We also measured association of these compounds to the more distantly related phosphatase SHP2.²⁵ Neither displayed significant binding at concentrations up to 100 μM. Therefore, as hoped, the strategy of seeking ligands that bind outside of the catalytic domain of phosphatases is indeed able to return selective ligands.

As mentioned above, IP⁺-containing macrocycles routinely display improved cell permeability over analogues lacking this unit.^{16, 17} To determine if this is the case here, cellular permeability was quantified using the chloroalkane penetration assay (CAPA), which measures cytosolic entry of chloroalkane-tagged compounds via HaloTag capture.²⁶ Titration of cells with a chloroalkane-tagged (Ct) compound for a set period of time allows permeability to be quantified, providing a CP₅₀, defined as the concentration of compound required to achieve 50% occupancy of the HaloTag protein. To evaluate the cellular permeability of Compounds 4 and 7, their chloroalkane-tagged derivatives (4-Ct and 7-Ct; Figure 6D) were synthesized and tested. Compounds 18-Ct and 19-Ct (structures shown in Figure 6D), chloroalkane-tagged macrocycles that structurally resemble 4-Ct and 7-Ct but lack the IP⁺ unit, were also tested. Chloroalkane-tagged JQ1 (JQ1-Ct, Figure 6D), a 457 Da, Lipinski-compliant²⁷ BRD4 inhibitor,²⁸ served as a high permeability benchmark.¹⁷ As shown in Figure 6E, Compound 4-Ct and 7-Ct exhibited CP₅₀s of 8 μM and 10 μM, respectively. In contrast, the control compounds 18 and 19, which lack the IP⁺ unit, displayed CP₅₀s > 50 μM, providing yet another example of the significant stimulatory effect of the IP⁺ ring on the passive movement of MPs across the cell membrane. The CP₅₀ of JQ1-Ct was found to be 0.4 μM, so we would categorize 4-Ct and 7-Ct as moderately cell permeable molecules, which is not surprising since no effort has been made to optimize their properties, for example through N-methylation and elimination of non-essential hydrogen bond donors.

We also evaluated the membrane permeability of compounds 4 and 7, as well as their analogues lacking the IP⁺ heterocycle (structures shown in Figure S7A), using the Parallel Artificial Membrane Permeability Assay (PAMPA). As shown in Figure S7B, compounds 4 and 7 exhibited −LogP_e values of 6.41 and 6.60 respectively. The non-IP⁺-containing analogues 27 and 28 traversed the membrane much more slowly, with −LogP_e values of 7.87 and 7.42 respectively. These results confirmed that our PTP1B ligands are membrane permeable.

We conclude from the above experiments that this iterative function-based screening platform is capable of providing selective, non-inhibitory, cell permeable ligands for a PTP.

Extension of the screening platform to targets without enzymatic activity

This study, and our previously published report,¹⁰ demonstrate that non-inhibitory ligands for phosphatase and E3 Ubiquitin ligase enzymes can be obtained using this function-based, on-resin screening format. As suggested above (see Scheme 1), we hope that this general screening format could be extended to other types of enzymes as well. Of course, many interesting targets are not enzymes or catalyze reactions that are not readily assayed in this format. We hypothesized that ligands for such proteins could be identified on our platform using proximity labeling technology,^{13, 29} specifically by screening against a fusion protein comprised of the target and a biotin ligase using beads with a co-displayed peptide substrate for this enzyme. Visitation of the fusion protein to the bead surface would result in biotinylation of the co-displayed substrate, which is readily detectable using labeled Streptavidin or anti-biotin antibody. Obviously, any hits identified in a screen against the fusion protein would have to be tested for association with the biotin ligase. We note that Krusemark and co-workers have employed a similar concept in DNA-encoded library screening.³⁰

To test this concept in the context of the bead system described here, we cloned, expressed and purified a fusion of Turbo ID,¹³ an engineered biotin ligase, with the PRU domain of Rpn13, one of the Ubiquitin receptors on the proteasome.³¹ The PRU domain has no known catalytic activity. It is a scaffolding protein that associates with the C-terminal region of Rpn2 as well as Ubiquitin chains.³² It is a target for the construction of Ubiquitin-independent degraders, which are chemical dimerizers that recruit a target protein directly to the proteasome.³³ As before, 10 μm TentaGel beads were functionalized such that 15% of the available surface sites were equipped with azide handles for subsequent substrate conjugation. The remaining 85% of modifiable sites were used for ligand immobilization (Figure 7A). Three reported Rpn2-derived peptides with Rpn13 PRU-binding affinities of 15 nM, 910 nM and 22 μM, respectively³² (Figure S8), were used as model ligands, while a “naked” PEG linker served as a negative binding control. An alkyne-modified AviTag peptide was conjugated to the azide handles via click chemistry to serve as the bead-displayed substrate for TurboID-mediated proximity labeling.

Figure 7.

Establishment and validation of a proximity-driven labeling screening platform using TurboID-PRU and Rpn2-derived peptides as the model target-ligand pair. (A) Schematic design of assay beads for on bead biotinylation studies. (B) Flow cytometry analysis of negative control beads after incubation with TurboID-PRU at the indicated concentrations and subsequent staining with Alexa Fluor 647-conjugated anti-biotin antibody. The corresponding beads displaying synthetically biotinylated AviTag were also analyzed to define the signal associated with complete biotinylation. (C) Flow cytometry analysis of negative control beads and beads displaying Rpn2-derived peptide 20, 21, or 22 after incubation with TurboID-PRU at the indicated concentrations and subsequent staining with Alexa Fluor 647-conjugated anti-biotin antibody. (D) Schematic design of assay beads co-displaying AviTag and an Rpn13-binding fragment. (E) Flow cytometry analysis of negative control beads, beads displaying Rpn2-derived peptide 20, and beads displaying an Rpn13-binding fragment after incubation with TurboID-PRU at the indicated concentrations and subsequent staining with Alexa Fluor 647-conjugated anti-biotin antibody.

We first assessed the level of ligand-independent biotinylation. Negative control beads lacking a ligand were incubated with the TurboID-PRU fusion protein at concentrations ranging from 25 nM to 1 μM for 16 hours at 37°C. Following termination of the incubation, beads were stained with an Alexa Fluor 647-conjugated anti-biotin antibody, and bead biotinylation levels were quantified by flow cytometry. Beads bearing synthetically biotinylated AviTag and the PEG negative control at the same ratio as the negative control beads were also stained with the same antibody to establish the positive signal control and benchmark the maximum on bead biotinylation that can be achieved. As shown in Figure 7B, negative control beads incubated with TurboID-PRU fusion protein at a concentration of 100 nM or lower exhibited biotinylation levels at least two orders of magnitude lower than those of the positive control.

We next benchmarked the sensitivity of this platform. ≈ 250K beads displaying one of the three Rpn13-binding peptides or the unsubstituted PEG linker, were individually incubated with the TurboID-PRU fusion protein at concentrations ranging from 25 nM to 1 μM at 37 °C for 16 hours, then stained with A647-labeled anti-biotin antibody and analyzed by flow cytometry. As shown in Figure 7C, even at Rpn13 PRU-TurboID concentrations as low as 25 nM, all ligand-displaying beads exhibited significantly higher biotinylation levels than the negative control beads, including even the beads displaying the lowest affinity peptide. Also as expected, the biotinylation levels of ligand-displaying beads correlated with the binding affinities of the displayed ligands. Increasing the fusion protein concentration further widened the signal gap between negative control beads and beads displaying low-affinity Rpn2-derived peptides, suggesting the assay is capable of registering ligands with K_D values significantly higher than 20 μM.

To test this hypothesis, we synthesized beads (structure shown in Figure 7D) that displayed a previously identified Rpn13 PRU-binding fragment with a K_D of only 2 mM.³⁴ Nonetheless, this 7-azaindole fragment supports a level of bead biotinylation readily distinguishable from that of the negative control (Figure 7E). As expected, the intensity of the signal was lower than that supported by the high affinity peptide.

Discussion

As evidenced by the work described above, we have established a powerful workflow for the discovery of non-inhibitory ligands of PTP enzymes, here using PTP1B as an interesting exemplar. The key element of this platform is the coupling of binding of the phosphatase to a bead-displayed ligand and proximity-driven modification of a co-displayed, high K_M substrate. This results in a stable, covalent mark of the binding event, which is readily detected using a fluorescently labeled anti-phosphotyrosine antibody. This coupling strongly biases the screen against the recovery of inhibitors.

We found that the assay is quite sensitive and capable of identifying even weak ligands. This, in turn, allowed us to employ a FBDD-inspired, step-by-step sequential screening approach (Figure 3) that eventually delivered a highly selective, 1 μM ligand for PTP1B through three rounds of screening using fewer than 300 compounds. Thanks to the presence of the IP⁺ unit, the macrocycle displayed moderate cell permeability, even though no effort was made to minimize hydrogen bond donors in the peptide. While we focused solely on PTP1B in this study, it seems likely that a similar approach would lead to macrocyclic ligands against other PTPs.

Two of the positions in macrocycle 7 appear to be solvent exposed and unable to contribute to PTP1B binding. This is evidenced by the fact that placement of bulky residues at these positions does not weaken the affinity for PTP1B. In addition, variation of the residues at position 3 failed to yield improved ligands. Thus, increasing the binding affinity will have to be done through optimization of the contacts made by the units at positions 1 and 2, as well as on the IP⁺ ring. This would be facilitated by structural information, which we are working to obtain.

The fact that only one “hemisphere” of macrocycle 7 engages PTP1B is perhaps an advantage with respect to the construction of PTP1B-recruiting PHORCs, since it provides a clear exit vector. It is also possible that libraries in which the remaining positions in 7 are diversified could provide molecular glues that recruit PTP1B and some other protein into a ternary complex. For example, one could employ our previously published assay¹⁰ to identify macrocycles able to recruit an E3 Ubiquitin ligase to the bead-displayed macrocycle-PTP1B complex, resulting in poly-Ubiquitylation of the phosphatase. A selective degrader would be of keen interest as a possible therapeutic for Type II diabetes,³⁵ since PTP1B opposes insulin signaling.³⁶

Finally, we also demonstrated here that this assay platform can be used to identify ligands for proteins that lack enzymatic activity by employing a TurboID fusion protein as the target and co-displaying on the resin an AviTag peptide. This proximity labeling approach was inspired by the work of Krusemark and co-workers who used it to aid in screening of soluble DNA-encoded libraries.³⁰ This platform represents an appealing general alternative to ligand discovery form one bead one compound (OBOC) libraries³⁷ using a simple binding assay based on non-covalent capture of a labeled enzyme by a bead-displayed ligand.

Supplementary Material

Full experimental details, materials and methods, supplementary figures and data, including a listing of the building blocks employed for library construction.