Chemical activation of divergent protein tyrosine phosphatase domains with cyanine-based biarsenicals

Strategies for the direct chemical activation of specific signaling proteins could provide powerful tools for interrogating cellular signal transduction. However, targeted protein activation is chemically challenging, and few broadly applicable activation strategies for signaling enzymes have been developed. Here we report that classical protein tyrosine phosphatase (PTP) domains from multiple subfamilies can be systematically sensitized to target-specific activation by the cyanine-based biarsenical compounds AsCy3 and AsCy5. Engineering of the activatable PTPs (actPTPs) is achieved by the introduction of three cysteine residues within a conserved loop of the PTP domain, and the positions of the sensitizing mutations are readily identifiable from primary sequence alignments. In the current study we have generated and characterized actPTP domains from three different subfamilies of both receptor and non-receptor PTPs. Biarsenical-induced stimulation of the actPTPs is rapid and dose-dependent, and is operative with both purified enzymes and complex proteomic mixtures. Our results suggest that a substantial fraction of the classical PTP family will be compatible with the act-engineering approach, which provides a novel chemical-biological tool for the control of PTP activity and the study of PTP function.

. Design of chemically activatable (act) PTPs. (A) In actPTPs, three cysteines are engineered at nonconserved residues within the WPD loop to create a tricysteine motif. Engagement of a biarsenical ligand with the engineered binding site on the WPD loop enhances actPTP activity while wild-type activity is unaffected. (B) Primary sequence alignment of the WPD loops of wild-type PTPs. In the WPD loop sequence of actPTPs, cysteines replace the positions colored in red. (C) Chemical structures of four biarsenicals used to target actPTPs.
and Src-homology-2-domain-containing PTP 2 (SHP2) 33 . HePTP (subtype R7, 38% PTP-domain identity with PTP1B) is expressed in white blood cells, negatively regulates T-cell activation and proliferation, and is overexpressed in some preleukemic myeloproliferative diseases 34 . PTPκ (subtype R2A, 37% PTP-domain identity with PTP1B) is a receptor PTP that has been implicated as a breast-cancer suppressor 35 . Downregulation of PTPκ has also been linked to melanoma, lung cancer, and prostate cancer 36 . SHP2 (subfamily NT2, 39% PTP-domain identity with PTP1B) provides another example of the strong connection between misregulation of PTP activity and human disease, as germline SHP2 mutations cause Noonan and LEOPARD syndromes [37][38][39] and activating somatic SHP2 mutations are the most common cause of sporadic juvenile myelomonocytic leukemia 40,41 .
To design putatively activatable mutants of HePTP, PTPκ, and SHP2, we aligned each enzyme's WPD-loop sequence with that of PTP1B and identified the amino-acid residues that correspond to the cysteine-mutation sites of actPTP1B (V184, E186, and S187 of wild-type PTP1B, Fig. 1B). Site-directed mutagenesis was used to place three cysteine residues at the corresponding WPD-loop positions to generate actHePTP and actPTPκ. To design actSHP2, we coupled the analogous tri-cysteine WPD loop mutation with the removal of a non-conserved cysteine (C333) at a different site on the enzyme's PTP domain, due to the previous observation that C333 renders SHP2 sensitive to inhibition by biarsenical compounds 42 . To avoid potential competition between biarsenical-induced activation and inhibition in the tri-cysteine engineered SHP2, we mutated C333 to proline, the residue that is found at the corresponding position in the vast majority of human classical PTP domains 33 . The three engineered proteins, actHePTP, actPTPκ, and actSHP2, expressed well in Escherichia coli and were isolated at purity levels indistinguishable from those of their wild-type counterparts.
Identification of AsCy5 as an activator of actHeptp. As a first test of the possibility of extending the actPTP approach beyond the PTP1B/TCPTP subfamily (NT1) we selected actHePTP. To determine if actHePTP is targetable by biarsenicals, we incubated the purified enzyme with four commercially available compounds: FlAsH, ReAsH, AsCy3, and AsCy5 (Fig. 1C). Activity assays using the substrate para-nitrophenylphosphate (pNPP) revealed that all four biarsenicals induce substantial activation of actHePTP ( Fig. 2A). In contrast to previous studies with actPTP1B and actTCPTP, on which AsCy3 was the strongest activator 31 , we found that AsCy5, a compound with a larger interarsenical distance, was the most effective activator of actHePTP. Pre-incubation with 500 nM AsCy5 gave rise to an approximately 175% increase in actHePTP's dephosphorylation rate ( Fig. 2A), but had no significant effect on the activity of wild-type HePTP ( Supplementary Fig. S1A). These results demonstrate that the tri-cysteine WPD-loop mutation strategy can be extended beyond the NT1 subfamily of PTPs, and that screening a panel of biarsenicals with varying interarsenical distances can be useful for optimizing the efficacy of biarsenical-induced activation when applying the strategy to new PTP subfamilies.
Characterization of AsCy5-induced activation of actHeptp. We next investigated the dose-dependence of AsCy5-induced activation of actHePTP. Following incubation of actHePTP (200 nM) with various concentrations of AsCy5, we observed potent dose-dependent activation that plateaued at approximately 750 nM AsCy5 and four-fold activation (Fig. 2B). At lower concentrations, relative activity increased steadily with increasing AsCy5; even the lowest concentration tested (31.25 nM) produced approximately 150% activity compared to the vehicle-only control. The remarkable potency of AsCy5-mediated activation is evidenced by the low compound concentration necessary to reach half maximum activation (AC 50 ≈ 200 nM AsCy5), which approaches the assay's concentration of enzyme in solution (200 nM actHePTP).
To assess the rate at which AsCy5 exerts its effect on actHePTP, we next measured the time dependence of activation. We incubated actHePTP with AsCy5 for various time periods and then tested for activity (Fig. 2C). AsCy5-induced activation was rapid, with nearly two-fold activation observed after only one minute of pre-incubation, and the assay's maximum activation was observed within ten minutes.
We next compared the kinetic parameters of the activated actHePTP with those of the non-treated enzyme ( Table 1 and Supplementary Fig. S2). We also directly compared the activity of untreated actHePTP with that of wild-type HePTP to assess the potential effect of the cysteine mutations on HePTP's inherent enzymatic activity. We found that catalytic efficiency (k cat /K M ) of actHePTP in the absence of biarsenical was roughly comparable to that of wild-type (Table 1). Although the catalytic rate constant (k cat ) of untreated actHePTP was slightly lower than that of wild-type HePTP, the small reduction in k cat was offset by a comparable reduction in Michaelis constant (K M ). In the presence of AsCy5, the catalytic efficiency of actHePTP increased dramatically, with a 1.7-fold increase in k cat and a 3.5-fold decrease in K M combining for a six-fold increase in k cat /K M . It follows from the data described above that AsCy5-treated actHePTP dephosphorylates pNPP much more efficiently than does wild-type HePTP, predominantly due to a 5.6-fold decrease in K M (Table 1). To our knowledge, AsCy5-treated actHePTP represents the first small-molecule modulator that is capable of increasing HePTP activity above that of the wild-type enzyme.
Activity of actHeptp with phosphopeptide substrates. For our engineered activation approach to be useful in the exploration of cellular pathways involving HePTP, actHePTP must be able to act on physiological substrates of HePTP, and AsCy5 must accentuate actHePTP activity regardless of substrate. To test both actHePTP's ability to dephosphorylate a HePTP substrate and the substrate-dependence of AsCy5-induced activation of actHePTP, we measured the enzyme activity on the phosphopeptide TGFLTE(pY)VATR, which corresponds with a phosphorylation site on the mitogen-activated protein kinase 1 (MAPK1), a physiological substrate of HePTP 43 .
In the absence of AsCy5, actHePTP dephosphorylates TGFLTE(pY)VATR at a rate that is notably lower than that of wild-type HePTP, but addition of AsCy5 induces activation of actHePTP slightly above wild-type levels (Fig. 2D). The augmentation of actHePTP's phosphopeptide dephosphorylation rate beyond that of wild-type HePTP is modest in comparison to the activation observed with pNPP as a substrate. However, these data importantly demonstrate that actHePTP is both active and activatable in the dephosphorylation of a model peptide based on the parent enzyme's physiological substrate.
To determine whether activation levels are consistent among varying phosphopeptides, we also measured the effect of AsCy5 on actHePTP activity on the phosphopeptide DADE(pY)LIPQQG, which corresponds to an autophosphorylation site on the epidermal growth factor receptor (EGFR), a physiological substrate of PTP1B 44 . In line with our observation on TGFLTE(pY)VATR, actHePTP dephosphorylates DADE(pY)LIPQQG at a rate that is notably lower than that of wild-type HePTP in the absence of AsCy5, but addition of AsCy5 induces activation of actHePTP slightly beyond wild-type levels (Fig. 2E). Broadly, the actHePTP AsCy5-induced activation results were largely consistent between the two phosphopeptides tested, demonstrating that AsCy5 is capable of direct chemical activation of actHePTP on a variety of substrates.
targeting actHePTP with AsCy5 in complex proteomes. We next asked whether actHePTP could be targeted for direct chemical activation in the presence of a complex proteome. To do so, we measured the dose-dependence of AsCy5-induced activation of actHePTP in a cell lysate derived from actHePTP-expressing Escherichia coli. (Because the E. coli genome does not encode PTPs, total PTP activity of a crude E. coli lysate can be interpreted as representative of the activity of an overexpressed mammalian PTP.) After incubating lysates from actHePTP-expressing cells with various concentrations of AsCy5 (or vehicle control), we assayed each www.nature.com/scientificreports www.nature.com/scientificreports/ mixture for PTP activity with pNPP (Fig. 3A). Similar to results on purified actHePTP, AsCy5-induced activation of actHePTP in a crude lysate was dose-dependent and robust, although activation in lysates required moderately higher AsCy5 concentrations (AC 50 ≈ 500 nM) than the activation of purified actHePTP (AC 50 ≈ 200 nM).
To investigate whether the need for higher AsCy5 concentrations was due to off-target binding within the crude cellular lysate, we took advantage of AsCy5's inherent fluorescent properties. After incubating crude lysates with various concentrations of AsCy5, we separated the lysate proteins by SDS-PAGE and visualized AsCy5-bound proteins with near-IR light. Even at the highest AsCy3 concentration (2000 nM), the lane derived from an AsCy5-treated wild-type HePTP-expressing lysate displayed no strong bands, whereas lanes derived from AsCy5-treated actHePTP-expressing lysates each displayed a strong band that is found at a position corresponding to the molecular weight of HePTP (35 kDa, Fig. 3B). This band diminishes in intensity with decreasing concentrations of AsCy5, consistent with the observed dose-dependent activation. These results suggest that AsCy5 engages with actHePTP with high specificity, even in a complex proteomic environment.
The ability of AsCy5 to cross membranes has not been investigated, but the compound's cell permeability is presumably hindered due to its two negatively charged sulfonate moieties, much like the structurally similar AsCy3 (Fig. 1C) 45 . However, we previously found that high concentrations of AsCy3 can be used to target actPTP1B in intact cells 31 , leading us to investigate AsCy5's ability to activate actHePTP in E. coli. We incubated actHePTP-expressing cell suspensions with various concentrations of AsCy5 and washed away free compound. After lysis and clarification, the resulting solutions were tested for PTP activity (Fig. 3C). The data suggest that AsCy5 has very limited cellular permeability, at least in bacteria, as the observed AsCy5-induced actHePTP activation after in-cell treatment was not nearly as potent or as robust as observed on purified enzyme and in membrane-free proteomic mixtures. It is likely that targeting of actHePTP in mammalian cells would require the development of AsCy5 analogs that lack the parent compound's sulfonate groups.
AsCy5-induced activation of actptpκ. We further explored the generalizability of our activation approach by extending our studies to the putatively activatable actPTPκ. We investigated the potential for biarsenical-induced activation of actPTPκ by incubating the enzyme with FlAsH, ReAsH, AsCy3, and AsCy5. Like actHePTP, actPTPκ activity was most robustly augmented by AsCy5 (approximately 425% activity), while the effects of FlAsH, ReAsH, and AsCy3 were more modest (Fig. 4A). We therefore selected AsCy5 for further characterization of actPTPκ activation.
To characterize the dose-dependence of AsCy5-mediated activation of actPTPκ we measured its activity following incubation with various concentrations of AsCy5. We observed dose-dependent activation of actPTPκ that reached AC 50 at approximately 600 nM and plateaued at 2000 nM AsCy5 (Fig. 4B). Although the dose-dependent activation of actPTPκ appears to be 2-3-fold less potent than the activation of actHePTP (compare Fig. 4B with 2B), the apparent discrepancy is most likely due to the different enzyme concentrations used (actHePTP: 200 nM, actPTPκ: 400 nM), as AsCy5-induced activation of both enzymes approaches the theoretical potency limits of the assays.
To assess the rate at which AsCy5 binds to actPTPκ's engineered tri-cysteine motif on the WPD loop, we measured the time dependence of activation. We incubated actPTPκ with AsCy5 for various time periods and then tested for activity (Fig. 4C). AsCy5-induced activation was rapid, with maximum activation observed after approximately sixteen minutes of pre-incubation.
Michaelis-Menten analysis of wild-type PTPκ and actPTPκ reveals that AsCy5-treated actPTPκ shows a five-fold increase in catalytic efficiency when compared to a vehicle-only actPTPκ control and a four-fold increase in catalytic efficiency when compared to wild-type PTPκ ( Table 2 and Supplementary Fig. S3). As with actHePTP, this increase is due to a combination of an increase in k cat and a decrease in K M . Additionally, AsCy5 successfully activates actPTPκ in the dephosphorylation of phosphopeptide DADE(pY)LIPQQG. The dephosphorylation of DADE(pY)LIPQQG by actPTPκ occurs at a rate that is notably lower than that of wild-type PTPκ in absence of AsCy5, but addition of AsCy5 induces activation of actPTPκ back to wild-type levels (Fig. 4D).
We next investigated AsCy5-induced activation of actPTPκ in E. coli lysates. The dose-dependent activation of actPTPκ in the proteomic mixture is remarkably similar to the dose-dependent activation of purified actPTPκ; maximum activity of actPTPκ within the crude lysate is approximately 325% at 2000 nM AsCy5 (Fig. 5A). Visualization of lysate proteins after SDS-PAGE separation and exposure to near-IR light shows one strong band in all lanes at a position that corresponds with the molecular weight of the PTPκ catalytic domain (34 kDa), with band intensity diminishing with decreasing AsCy5 concentration (Fig. 5B). Unexpectedly, we found that a high concentration of AsCy5 (4000 nM) also produced detectable labeling of the wild-type PTPκ catalytic domain (Fig. 5B), even though wild-type PTPκ contains no engineered tri-cysteine motif. Closer inspection of PTPκ's primary and tertiary structure revealed that, in contrast to the other PTPs studies here, the C-terminal portion of PTPκ's catalytic domain contains three cysteine residues in close proximity to one another (C1143, C1145, and C1153, Supplementary Fig. S4) 46 . This non-conserved cysteine-rich motif of wild-type PTPκ likely constitutes the AsCy5-binding site suggested by the data in Fig. 5B. Importantly, PTPκ's naturally occurring cysteine-rich motif is not near the enzyme's active site, and we found that the activity of wild-type PTPκ is neither activated nor significantly inhibited by AsCy5 (Supplementary Fig. S1B).
AsCy3-induced activation of actSHP2. We next investigated the biarsenical-induced activation of act-SHP2, by measuring its activity after pre-incubation with FlAsH, ReAsH, AsCy3, and AsCy5 (Fig. 6A). Both cyanine-derived biarsenicals, AsCy3 and AsCy5, were more successful in activating actSHP2 than FlAsH and ReAsH. AsCy3 was the most robust activator (approximately 550% activity), and was selected for further characterization.  www.nature.com/scientificreports www.nature.com/scientificreports/ Upon testing the dose-dependence of AsCy3-mediated activation of actSHP2, we observed robust and potent activation that plateaued between 500 nM and 1000 nM AsCy3 at approximately 800% activity (Fig. 6B). We then incubated actSHP2 with AsCy3 for various time periods to assess the rate at which AsCy3 acts on the engineered tri-cysteine motif of the WPD loop (Fig. 6C). AsCy3 binds quickly, with over 300% activity observed at 20 seconds and maximum activation observed after approximately six minutes. Consistent with previous findings that the presence of the non-conserved cysteine C333 renders SHP2 sensitive to inhibition by biarsenicals, we found that wild-type shows some inhibition in the presence of 1000 nM AsCy3 ( Supplementary Fig. S1C), validating the previously described engineering strategy of introducing the C333P mutation into actSHP2.
Michaelis-Menten analysis of wild-type SHP2 and actSHP2 reveals that the activity of actSHP2 is virtually indistinguishable from that of wild-type SHP2 in the absence of biarsenical. AsCy3-treated actSHP2 shows a remarkable ten-fold increase in catalytic efficiency due, again, to the combination of an increase in k cat and a decrease in K M (Table 3 and Supplementary Fig. S5). This AsCy3-induced activation of actSHP2 persists to a lesser extent in the desphosphorylation of phosphopeptide DADE(pY)LIPQQG. The dephosphorylation of phosphopeptide DADE(pY)LIPQQG by actSHP2 occurs at a slightly lower rate than dephosphorylation by wild-type SHP2 (Fig. 6D). Addition of AsCy3 restores the activity of actSHP2 back up to wild-type levels.
Sensitivity of actptps can be governed by the residue preceding the act motif. To explore potential application of the act-engineering strategy across multiple members of a single PTP subfamily, we selected PTPRR, a second member of HePTP's subfamily, R7 33 . Based on the actHePTP prototype, we generated the putatively activatable actPTPRR through introduction of a WPD-loop tricysteine motif (Fig. 7A). In contrast to previous examples of actPTPs, however, actPTPRR possesses catalytic activity that is notably reduced from that of wild-type PTPRR (Table 4 and Supplementary Fig. S6). We further found that, although actPT-PRR is technically activatable by biarsenicals (approximately 300% activity with ReAsH and AsCy3, Fig. 7B), its biarsenical-induced rate enhancements are relatively modest when compared to its 13-fold defect in inherent catalytic efficiency ( Table 4).
Inspection of actPTP WPD-loop sequences outside of the cysteine-mutation sites revealed a potential reason for the failure of the act strategy when applied to PTPRR. PTPRR contains a lysine residue at position 556, a position that lies adjacent to the first cysteine residue of the act motif and is occupied by glycine in the vast majority of human PTP domains (Fig. 7A) 33 . PTPRR is one of only two PTPs that have a lysine at this position, and the actPTPs that have been the most robustly responsive to biarsenicals in this and previous studies (e.g., actPTP1B, actTCPTP, actSHP2) 31 Table 2. Kinetic constants of wild-type (wt) PTPκ, actPTPκ, and AsCy5-treated (4 µM) actPTPκ assayed with pNPP. www.nature.com/scientificreports www.nature.com/scientificreports/ www.nature.com/scientificreports www.nature.com/scientificreports/ structure (Fig. 7C) 47 . Although K556's role, if any, in favoring the intermediate WPD-loop structure is unknown, the sequence and structural idiosyncrasies of the PTPRR WPD loop, coupled with actPTPRR's low responsiveness to biarsencials, led us to hypothesize that the presence of the PTP-consensus residue, glycine, at position 556 may be necessary for strong activation (>500%) of actPTPs.  Table 3. Kinetic constants of wild-type (wt) SHP2, actSHP2, and AsCy3-treated (1 µM) actSHP2 assayed with pNPP. www.nature.com/scientificreports www.nature.com/scientificreports/ To test this hypothesis we generated K556G actPTPRR (Fig. 8A) and tested its activity in the absence and presence of biarsenicals. Although the glycine mutation at position 556 did not rescue the inherent catalytic defect of actPTPRR, as shown by K556G actPTPRR's low catalytic efficiency ( Supplementary Fig. S7), activation of K556G actPTPRR with the biarsenical AsCy3 was found to be robust (Fig. 8B). The approximately 900% activity of K556G actPTPRR that is induced by AsCy3 is on par with, or exceeds, the strong activation observed on actPTPs that naturally contain glycine at this position (actPTP1B, actTCPTP, actSHP2, and actPTPκ) 31 . These results strongly suggest that the presence of the WPD-loop's consensus glycine residue is essential for optimal biarsenical-induced activation of act-engineered PTPs, and that future applications of actPTP strategy will find the strongest success with PTPs that naturally possess the consensus residue. Since 31 out of 37 human classical PTPs do have glycine at this position, however, the "glycine criterion" constraint should not present a major impediment to the prospect of broadly applying the actPTP strategy.

conclusions
Although small-molecule-based protein activation strategies could provide powerful tools for probing cell-signaling pathways, few such methods have been described. Here we have shown that sequence-guided engineering of the WPD loop of classical PTPs provides a highly generalizable means to render PTPs activatable by the cyanine-based biarsenicals AsCy3 and AsCy5. We have shown that actPTPs can be systematically generated from various subfamilies of both receptor and non-receptor PTPs and that biarsenical-induced stimulation of actPTPs is rapid, dose-dependent, and potent, even in the context of complex proteomic mixtures. Although one specific WPD-loop sequence requirement for successful actPTP engineering was also outlined in this study, the limitation presented by this requirement is minor, as the majority of classical PTPs contain the key sequence element. In sum, our findings establish the broad generalizability of actPTP engineering and suggest that a substantial fraction of the classical PTP family will be compatible with the approach, which provides a novel tool for the small-molecule-based induction of PTP activity and the study of PTP function.

Methods
General and materials. "% PTP Activity" is defined as enzymatic reaction rate in the presence of a biarsenical as a percentage of the rate of a corresponding DMSO control (100%). Error bars and "±" values represent the standard deviation of at least three independent measurements. Biarsenicals were obtained commercially and used without further purification.
cloning and mutagenesis of ptp-encoding genes. The plasmid encoding the His 6 -tagged catalytic domains of SHP2 has been previously described 42 . Plasmids for expression of the His 6 -tagged catalytic domains of HePTP (pET-HePTP, residues 65-360), PTPκ (pDK002, residues 870-1154), and PTPRR (pBAP004, residues  Table 4. Kinetic constants of wild-type (wt) PTPRR and actPTPRR, assayed with pNPP.  phosphatase activity assays with purified enzymes. Quenched phosphatase activity assay using pNPP. Quenched activity assays using pNPP as substrate were performed as previously described 31  Continuous phosphatase activity assay with pNPP. Continuous activity assays with pNPP were carried out as previously described 31 . Briefly, after pre-incubations of varying times (see figures) in PTP buffer (see above), reactions were started by the addition of pNPP (at varying concentrations) to solutions containing PTP (at varying concentrations) and biarsenical (at varying concentrations) or DMSO. The absorbances (405 nm) of the resulting solutions were measured continuously.
Continuous phosphatase activity assay with phosphopeptide substrate. PTP kinetic assays with phosphopeptides DADEpYLIPQQC and TGFLTEpYVATR were carried out by measuring increasing absorbance at 282 nm, essentially as described 31 . Assays were performed at 22 °C in peptide buffer (50 mM 3,3-dimethylglutarate at pH 7.0, 50 mM NaCl) with biarsenical (varying concentrations) or DMSO (1%) and the appropriate PTP (varying concentrations). Reactions were started by the addition of phosphopeptide (varying concentrations: see figures).
PTP assays with biarsenical-treated crude cell lysates. PTP assays with biarsenical-treated crude cell lysates were performed essentially as previously described 31 . E. coli overexpressing the appropriate PTP were grown and lysed. Protein concentrations of the clarified lysates were normalized, and lysate PTP activities were assessed as described above (quenched pNPP assay). Total protein concentration in the PTP assays varied between PTPs (see figures).
PTP assays with biarsenical-treated E. coli cells. PTP assays with biarsenical-treated E. coli cells were performed essentially as previously described 31 . Briefly, BL21(DE3) E. coli cells overexpressing the appropriate PTP were grown and harvested, resuspended, and incubated with either biarsenical (varying concentrations) or DMSO for 2 hours. After incubation, the cells were pelleted, washed and lysed. PTP activity in the clarified lysates was determined as described above (quenched pNPP assay).

SDS-pAGe and visualization of biarsenical-binding proteins.
Lysates from E. coli overexpressing the appropriate PTP were prepared as described above. Biarsenicals (varying concentrations) were added to lysate solutions (total protein concentrations vary: see figures). After incubation, 15 µL of diluted lysate proteins were separated by SDS-PAGE gel and visualized with a focus offset of 1.5 mm using the 700 nm laser of the Li-Cor Odyssey CLx Near-Infrared Imaging System. PageRuler Prestained NIR Protein Ladder was loaded for the determination of protein molecular weights.

Data availability
Supporting data is available from the authors.