PRL3 pseudophosphatase activity is necessary and sufficient to promote metastatic growth

Phosphatases of regenerating liver (PRLs) are markers of cancer and promote tumor growth. They have been implicated in a variety of biochemical pathways but the physiologically relevant target of phosphatase activity has eluded 20 years of investigation. Here, we show that PRL3 catalytic activity is not required in a mouse model of metastasis. PRL3 binds and inhibits CNNM4, a membrane protein associated with magnesium transport. Analysis of PRL3 mutants specifically defective in either CNNM-binding or phosphatase activity demonstrate that CNNM-binding is necessary and sufficient to promote tumor metastasis. PRLs they pseudo pseudophosphatases. formation phosphocysteine, which blocks CNNM-binding and may play a regulatory role. We show levels of PRL cysteine


Introduction
PRLs (phosphatase of regenerating liver) are small single-domain phosphatases within the family of protein tyrosine phosphatases (PTP). The first member, PRL1, was identified based on its increased expression in regenerating liver and mitogen-treated cells (1). All three phosphatases are highly overexpressed in tumors and promote growth, motility and cellular transformation both in culture and animal models (2)(3)(4). They are highly similar to each other (78 to 88% identity) and prenylated at their C-terminus, which targets them to plasma and endosomal membranes (5)(6)(7). PRL1 crystallizes as a trimer and trimers of all three PRLs have been detected in solution by crosslinking (8,9). Catalysis by PTP phosphatases proceeds via a two-step mechanism involving transfer of the phosphate to the enzyme and subsequent hydrolysis to regenerate the active enzyme (10,11). PRLs are unusual in that the second step is greatly slowed leading to the accumulation of the phosphocysteine intermediate both in in vitro and in vivo (12,13). Uniquely among PTPs, the phosphocysteine form of PRLs has a lifetime of hours in vitro. The result is that PRLs exhibit burst kinetics with very low steady-state catalytic activity (8,12). PRLs are also unusually sensitive to oxidation through formation of a disulfide bond between the catalytic cysteine and a neighboring cysteine residue, which further reduces their catalytic activity (8,12,14). Oxidation is a complex and relatively poorly understood regulatory mechanism of many protein phosphatases (15,16).
The mechanism of PRL oncogenicity remains an enigma. A number of different signaling pathways have been implicated but without a clear consensus (17)(18)(19). As an alternative, PRLs were shown to directly bind and inhibit a class of Mg 2+ transport associated proteins, CBS domain divalent metal cation transport mediators (CNNMs) (4,20). CNNMs are membrane proteins with a transmembrane domain that transports Mg 2+ and a cytosolic domain that binds PRLs (21). Binding is mediated by a conserved CNNM aspartic acid that acts as a pseudosubstrate and inserts into the PRL active site (13,22,23). There are four CNNM proteins in humans, which bind with similar affinity to all three PRLs. PRL binding inhibits CNNM-associated Mg 2+ efflux and acts to increase intracellular Mg 2+ (4,20,24). CNNM mutations are associated with hereditary diseases in humans: Jalili syndrome (25,26) and dominant hypomagnesemia (27). Studies in mice have identified roles in magnesium absorption, fertility, blood pressure, and tumorigenesis (4,(28)(29)(30). In C. elegans, loss of cnnm-1 and cnnm-3 cause a defect in gonadal development, which could be rescued by supplementing the media with Mg 2+ (31). And, in Drosophila, PRL1 and the fly CNNM homolog, Uex, fulfill a neuroprotective function in response to CO2 stimulation (32).
A very large number of studies have established the importance of the PRL catalytic site for cellular functions and oncogenicity; however, none distinguish between the phosphatase activity and CNNMbinding. Cysteine-to-serine or alanine mutations disrupt PRL inhibition of CNNMs (4) but they also abrogate phosphatase activity (12). Similarly, inhibitors that inhibit PRL cancer proliferation block both phosphatase activity and CNNM binding (33). Consequentially, the relative importance of PRL phosphatase activity versus CNNM binding has remained unknown. To address this, we designed PRL mutations that are specifically defective in one function and tested their activity in Mg 2+ efflux assays and an animal model of tumor metastasis. The results demonstrate that phosphatase activity is dispensable and highlight the importance of the PRL-CNNM signaling axis in cancer.

Mutagenesis of the PRL3 active site
Previous studies showed that substitution of the PRL3 catalytic cysteine by serine or alanine leads to loss of both phosphatase activity and CNNM binding (12,22). In protein phosphatases, the active site cysteine is present as a negatively charged thiolate so we reasoned that substitution of cysteine by aspartic acid might preserve CNNM-binding. To assess this, the PRL3 C104D mutant was prepared and its binding affinity measured by isothermal titration calorimetry (Fig. 1). The substitution of the catalytic cysteine by aspartic acid retained full binding while all other substitutions decreased binding affinity by 16fold or more. The permissiveness of the binding site was very restricted: substitution by either asparagine, which is isosteric with aspartic acid, or glutamic acid, which is isoelectric, strongly decreased binding. To validate this remarkable result, we generated the same mutations in the other two PRL phosphatases. Both PRL2 C101D and PRL1 C104D showed near wild-type affinity in ITC experiments (Fig. 1B & Suppl. Fig. S1). In contrast, mutations in the catalytic site (R101E) or the CNNM-binding site (R138E) blocked binding.
We next tested the ability of the PRL mutants to bind to CNNM proteins in cells. COS7 cells were transfected with Myc-tagged PRL3 mutant constructs and FLAG-tagged CNNM4. Proteins were subjected to coimmunoprecipitation analysis to investigate their interaction in cells. In agreement with the ITC results, only wild-type PRL3 and the C104D mutant associated with CNNM4 ( Fig.  1C).
Having identified a mutant that retained CNNM-binding, we turned to assays of PRL3 phosphatase activity. We performed two types of assays to assess PRL3 enzymatic activity. In the first assay, we used a synthetic fluorogenic substrate, DiFMUP, to measure the rate of phosphate hydrolysis. The assays were conducted under single turnover conditions to detect the initial rate of catalysis ( Fig. 2A). As expected, wild-type PRL3 showed good activity while the C104D and other cysteine substitutions were completely inactive. In contrast, mutation of arginine 138, which is part of the CNNM-binding site but outside of the catalytic site, had little effect on phosphatase activity. The R138E mutant displayed nearly wild-type activity (Fig. 2C).
As a second measure of phosphatase activity, we examined formation of the phosphocysteine intermediate upon incubation with a small molecule substrate, OMFP (22). As expected, Phos-tag SDS-PAGE analysis showed formation of the phosphocysteine intermediate upon incubation with a small molecule substrate by wild-type PRL3 and the R138E mutant while the cysteine mutants were inactive (Fig. 2B). Thus, we have identified two PRL3 mutations, C104D and R138E, that are selectively defective in either phosphatase activity or CNNM binding.

Crystal structures of mutant PRL complexes
The requirement for the PRL catalytic cysteine in CNNM binding is poorly understood. It does not make direct contacts with CNNM but its mutation has a large effect on binding. Crystal structures of PRL-CNNM complexes show only minor conformational changes despite large differences in affinity (13,22). Similarly, disulfide bond formation with an adjacent cysteine leads to a ~500-fold loss of affinity with only modest structural changes (22,23).
To understand the molecular basis for CNNM binding by the C104D mutant, we carried out structural studies using X-ray crystallography. We were able to obtain crystals of the cysteine-to-aspartic acid mutants of PRL1 and PRL2 in complexes with the CBS-pair domain of CNNM3 and CNNM2 (Fig. 3A). The mutant complexes adopted the same arrangement as the wild-type complex with binding mediated by an extended loop from CNNM3 that inserts into the PRL catalytic site (Fig 3B and Suppl. Fig. S2). Surprisingly, the substitution of cysteine by aspartic acid shifted the binding site residues by less than 0.5 Å.
Calculation of the partial charges in the PRL active site revealed the role of the catalytic cysteine in CNNM binding is primarily electrostatic (Fig. 3C). To act as a nucleophile, the pKa of the cysteine is shifted so it is deprotonated at neutral pH (10,11). Measurements with Yersinia PTP show combined the effects of the helix dipole and surrounding backbone amide groups shift the pKa by three pH units (34,35). We used the computational tool H++ to estimate the pKa of the PRL2 active site cysteine. H++ uses atomic coordinates to calculate protonation states based on the AMBER force field with an implicit continuum solvent model (36). The active site cysteine of PRL2 was calculated to have pKa of 4.5 to 5. Thus, the C104D mutation preserves high affinity binding of CNNM proteins because it conserves the negative charge of the cysteine thiolate (Fig.  3C). This paradoxical result is likely due to polarization of other PRL residues in the CNNM-binding site by the negative charge.
The effects of mutations could also be read out as changes in the thermal stability of the PRL proteins. We used differential scanning fluorimetry (thermal shift assays), which measures binding of a hydrophobic dye to the denatured protein, to measure the melting temperatures of PRL1, PRL2, and PRL3 ( Fig.  3D and Suppl. Fig. S3). For all three, the cysteine-to-aspartic mutation increased the melting temperature by 8 ºC or more, a large and highly significant change. The increase reflects the potential energy required to charge the cysteine. Strikingly, the PRL3 C104N mutation, which is the substitution of a single oxygen by nitrogen, led to a 22 ºC decrease in stability relative to C104D. The importance of charge over steric fit was confirmed by the C104E mutation, which led to a small increase in stability of 2 ºC. The R138E mutation that specifically inhibits CNNM binding had a negligible effect on protein stability.

Functional assay of mutant PRLs
We next turned to assessing the function of the PRL mutants in HEK293 cells transfected with Myc-tagged constructs. As controls, we analyzed cell extracts by immunoblotting to measure expression levels and immunofluorescence microscopy to confirm the mutants were properly localized. All the proteins expressed well (Fig. 4A) but the C104N mutant had a defect in its localization. Wild-type PRL3 and other mutants mainly localized at the plasma membrane but the C104N mutant formed aggregates in the cytosol (Fig. 4B). Accordingly, the C104N mutant was omitted from further analysis. PRL3 binding to CNNM proteins inhibits their Mg 2+ efflux activity but it is not known if the effect requires phosphatase activity (4,13,28). To address this, we tested the PRL3 mutants in a Mg 2+ efflux assay. HEK293 cells were transfected with CNNM4 and intracellular Mg 2+ levels monitored using the fluorescent indicator Magnesium Green (Fig.  4C). In the absence of PRL3, the Mg 2+ level dropped when Mg 2+ was removed from the extracellular milieu. Co-expression of wildtype PRL3 and the C104D mutant blocked the Mg 2+ efflux. Conversely, co-expression of mutants that prevent CNNM binding had no effect on Mg 2+ efflux. This demonstrates that the inhibition of CNNM activity by PRL3 is not due to dephosphorylation of CNNM4 or any other substrate.

PRL oncogenicity
We next addressed the activities of the PRL3 mutants in a well-established experimental model of tumorigenesis. In this model, B16 mouse melanoma cells are injected into the tail vein of mice and the number of lung metastases counted. Multiple groups have shown the importance of PRL3 in tumor implantation and outgrowth (4,37-39). Funato and colleagues showed that the C104S mutant is inactive in promoting metastases (4).
The new PRL mutants were tested to assess the relative importance of phosphatase activity versus CNNM binding. We verified the expression levels of the Myc-tagged PRL3 wild-type and mutant proteins by western blotting prior to injection (Fig. 5A). Levels were very similar with the possible exception of the C104E mutant, which showed a slightly less protein. Following injection and outgrowth, mouse lungs were excised, and the number of tumor nodules counted. B16 cells expressing wild-type PRL3 efficiently metastasized to the lung forming an average 60 nodules per animal, while the GFP-expressing control cells formed relatively few nodules ( Fig. 5B & C). Comparison of the C104D and WT PRL3 cell lines showed no significant difference in the number of lung nodules demonstrating that phosphatase activity is dispensable for metastasis. Conversely, the cell lines expressing the C104E and R138E mutant proteins, which are unable to bind CNNM proteins, were indistinguishable from the negative control. These results unambiguously show that PRL3 functions as a pseudophosphatase in this tumor model.

PRL phosphorylation
Unlike true pseudophosphatases, which have no catalytic activity (40), PRLs are functional enzymes and formation of the phosphocysteine intermediate prevents CNNM binding (13). To address the question of whether PRLs are constitutively active in binding CNNMs, we examined levels of PRL phosphorylation in cultured cells and tissues. Extracts were analyzed by Phos-tag SDS-PAGE gels and western blotting (Fig. 6A). As PRL3 expression is low and difficult to detect, we used a commercial antibody against PRL2 that also detects PRL1. Migration of phosphorylated proteins is retarded in Phos-tag gels and allowed us to assess the amount of phosphorylated and unphosphorylated protein.
Phosphorylation on cysteine can be specifically recognized, as it is sensitive to boiling prior to gel electrophoresis (13).
Analysis of an extract from HeLa cells grown with Mg 2+ showed PRL1 and PRL2 were predominantly phosphorylated. Upon transfer to a growth medium without Mg 2+ , there was a slow loss of the phosphorylated PRLs and an accumulation of the unphosphorylated forms. After 24 h, the majority of PRL1 and PRL2 were unphosphorylated. Upon re-addition of Mg 2+ to the culture medium, PRL1 and PRL2 were rapidly phosphorylated, revealing a burst of phosphatase activity. After one hour, the large majority of PRL1/2 was phosphorylated, roughly in the same proportions as prior to Mg 2+ withdrawal. In addition to changes in phosphorylation, Mg 2+ deprivation led to a marked increase in PRL levels, which is visible in both the Phos-tag and regular SDS-PAGE gel. The upregulation is the result of both transcriptional and translational regulatory mechanisms (41,42).
We next examined PRL phosphorylation in mouse tissues (Fig. 6B). As previously reported for PRL1 (43), protein levels were high in the colon with lower amounts in heart, liver and skeletal muscle. Comparison of boiled and unboiled samples and Phos-tag and regular SDS-PAGE gels showed that PRL1 and PRL2 were cysteine phosphorylated in all the tissues examined. The sample from the colon showed the most unphosphorylated protein; in the other tissues, the majority of PRL1 and PRL2 were phosphorylated. Taken together, these results show large changes in the proportion of PRLs bound to CNNMs in different culture conditions and tissues.

Uniqueness of PRL cysteine phosphorylation
The unusual kinetics of PRL phosphatase raised the question if other PTPs might similarly form stable phosphocysteine intermediates. To test this, we purified the catalytic domains of six phosphatases that have been previously reported to have very low activity (44). The domains were incubated with a synthetic substrate, OMFP, to generate the phosphocysteine intermediate and analyzed on Phos-tag gels with and without boiling to hydrolyze phosphocysteine (Fig. 7). Human PRL2 and PRL from a tardigrade species both showed formation of a slower migrating band upon incubation with OMFP. As previously observed, recombinant human PRL2 purified from E. coli was partially phosphorylated prior to OMFP treatment (13). None of the other phosphatases showed the formation of a stable phosphocysteine intermediate. These results suggest that stable cysteine phosphorylation is a unique and remarkably conserved property of PRLs.

Discussion
We have shown that PRL3 functions as a pseudophosphatase in the promotion of B16 tumorigenesis. PRL3 binds to CNNM proteins, which act as tumor suppressors by decreasing intracellular Mg 2+ levels (4,28). Mg 2+ has long been suspected to play a central role in coordinating control of metabolism and growth in animal cells (45). While Mg 2+ does not undergo the dramatic, multifold changes in intracellular concentration that Ca 2+ does, changes in Mg 2+ levels have been reported in response to cellular stimulation and growth signals. Most Mg 2+ is bound to phosphatecontaining compounds: CNNM proteins themselves selectively bind Mg 2+ ·ATP via their CBS-pair domain (24) and are important for regulating metabolism and preventing overproduction of reactive oxygen species (46). Compartmentalization of Mg 2+ could increase the size of the concentration changes and allow coordinated control of different pathways. PRL and CNNM proteins may also act through other downstream effectors.
As PRL3 does have phosphatase activity, in some sense, it should be considered as pseudopseudophosphatase. Although dispensable for PRL3 inhibition of CNNM Mg 2+ efflux and B16 tumorigenesis, it is unlikely that the catalytic activity of PRLs is without function. The PRL catalytic residues are conserved in all three isoforms and across metazoan species. The stability of the phosphocysteine form of tardigrade protein implies PRL cysteine phosphorylation has existed since the separation of protostomes and deuterostomes over 500 million years ago. PRL phosphorylation prevents CNNM binding and likely regulates CNNM function in cells (13). The tissue specific differences and changes in response to Mg 2+ availability argue for a conserved, physiological function of phosphatase activity, most likely as a mechanism for regulating the PRL-CNNM interaction.
B16 cells expressing the C104D mutant that cannot be phosphorylated might be expected to form more tumor nodules than cells expressing wild-type PRL3 (Fig. 5). We did not measure cysteine phosphorylation in the B16 cell lines but previous studies with exogenous expression of PRLs in HEK293 and COS7 cells showed high levels of PRL phosphorylation (13). One possible explanation for why the PRL3 C104D mutant does not act as a super oncogene is that the levels of wild-type PRL3 are high enough that the pool of unphosphorylated protein is sufficient to bind and inhibit CNNM proteins. Alternatively, some other aspect of migration, implantation, or growth may be limiting the number of tumor modules formed.
While PRL3 has significant initial (burst) phosphatase activity (Fig. 2), the long lifetime of the catalytic intermediate makes it an extremely poor enzyme under steady-state conditions with a turnover rate on the order of once per hour (12,13). As the second step in the catalytic cycle is rate-limiting, the catalytic throughput is not substrate dependent but depends on the rate of hydrolysis of phosphocysteine. While a cellular factor, protein or small molecule, could in principle increase the rate of phosphate release, rapid dephosphorylation of PRL1 and PRL2 was not observed in cells (Fig. 6). In response to removal of Mg 2+ from the culture medium, the levels of PRL phosphorylation decreased slowly, suggesting the absence of a mechanism for rapid dephosphorylation. In contrast, readdition of Mg 2+ led to rapid rephosphorylation. The substrates responsible for this rephosphorylation are not known but the observation of cysteine phosphorylation in PRL2 expressed in bacteria suggests broad substrate specificity (Fig. 7). Indeed, in vitro incubation of PRL2 with small phosphatecontaining compounds leads to partial cysteine phosphorylation (22). As PRL phosphatase activity is inhibited when bound to CNNM (13), one possibility is that phosphorylation acts as a switch to prevent re-association of PRLs following release. The dissociated, phosphorylated forms could mediate proposed non-catalytic functions of PRLs in TGFb and SOX2 signaling (47).
There is a great deal of interest in identifying specific inhibitors of PRLs as therapeutics against metastatic cancer (18,48,49). The majority of efforts have focused on inhibiting PRL catalytic activity and identifying substrates. Among the more successful compounds, thienopyridone inhibitors have been developed that take advantage of the sensitivity of the PRL catalytic cysteine to oxidation (33,50,51). Other approaches have identified allosteric inhibitors that disrupt PRL trimerization (52) and PRL-specific antibodies (53). Intriguingly, an inhibitor of the low molecular weight phosphatase LMPTP (ACP1) was identified that specifically binds and stabilizes the cysteine phosphorylated form of the protein (54). Thus, it may be possible to design small molecules that bind phosphorylated PRLs and alter the rate of phosphocysteine hydrolysis.
The identification of the C104D mutant should help these drug development efforts. The C104D mutant is insensitive to oxidation, which can lead to false positives in highthroughput screens. PRL3 C104D binding to CNNMs is not inhibited by redox active compounds (33). Additionally, the C104D mutant cannot be phosphorylated so it is constitutively active in binding and inhibiting CNNMs.
More generally, cysteine-to-aspartic acid substitutions may be useful as a tool for studying other enzymes with active site cysteines. In cysteine proteases and protein phosphatases, cysteine has a shifted pKa to act as a nucleophile so aspartic acid is the amino acid most closely matched in size and charge. The cysteine-to-aspartic acid mutations may be useful for substrate trapping (55) and offer the additional advantage of increased thermal stability. Indeed, the phosphatase PTPRG contains a pseudophosphatase domain with a substitution of aspartic acid for the catalytic cysteine (56). Although ionized cysteines are a relatively small proportion of cysteines in proteins, the general guidance that serine and alanine are most likely to be neutral mutations needs to be qualified.
In conclusion, we have identified and characterized a PRL3 phosphatase mutant that does not have catalytic activity but is oncogenic in a well-established animal model. The reciprocal mutant that is unable to bind CNNM is not oncogenic. These results unequivocally establish that PRL3 acts as a pseudophosphatase independently of its phosphatase activity.

Experimental procedures Expression and purification of proteins.
Human phosphatases PRL3 (residues 1-169), and PRL2 (residues 1-167) were expressed and purified as His-tag fusion proteins as described previously (13,22). Mouse phosphatase PRL1 (residues 7-160) was codon-optimized for E. coli expression and inserted into the pET15b vector (Novagen Inc., Madison, WI) modified to contain noncleavable MHHHHHH tag fusion protein at the N-terminus. Catalytic cysteine point mutants were generated with QuikChange Lighting Site-Directed Mutagenesis Kit (Agilent Technologies) and confirmed by DNA sequencing and mass spectrometry. Plasmids 38891 coding for PTPRN2 (715-1010) and 38944 coding for PTPN14 (886-1187) were obtained from Addgene (Cambridge, MA). DNAs coding for PTPRC (627-1228), PTPRM (867-1452), PTPN21 (875-1174), PTPRN (681-979), and PRL homolog RvY_14574 from Ramazzottius varieornatu were chemically synthesized (Bio Basic Inc., Markham, Canada) and cloned into pET29a (EMD Biosciences). The CBS-pair domain of human CNNM3 was prepared as previously described (13). The murine CNNM2 CBS-pair domain (residues 430-580) was cloned into the pET SUMO vector to produce a protein fused with His-tag and SUMO tag. The tags were removed using ULP-1 protease. The proteins were expressed in Escherichia coli strain BL21 and purified using affinity chromatography. After His-tagbased purification and tag cleavage for CBS domains, all proteins were additionally purified using Superdex-75 (GE Healthcare) size-exclusion column equilibrated with HPLC buffer containing 20 mM HEPES, 100 mM NaCl, 5 mM TCEP (tris(2carboxyethyl)phosphine), pH 7.0. PRL concentrations were estimated by absorbance at 280 nm using an extinction coefficient of 19940 M -1 cm -1 . Cell culture, transfection, and immunofluorescence analysis. COS7 and HEK293 cells were cultured in DMEM supplemented with 10% FBS and antibiotics. Expression plasmids were transfected in each cell with Lipofectamine 2000 (Invitrogen). Studies of PRL3 localization were performed as previously described for studies of CNNM4 localization (28) using a rabbit anti-Myc antibody (Santa Cruz) diluted in blocking buffer for Myc-PRL3 visualization and rhodamine-phalloidin (Invitrogen) for F-actin visualization. Cells were imaged with a FLUOVIEW FV1000 confocal scanning laser microscope (Olympus).
Phos-tag gels. For detection of protein phosphorylation, we used a Tris-tricine SDS-PAGE system with 20-40 µM Phos-tag reagent (Wako Chemicals, Japan) and 40-80 µM MnCl 2 added before polymerization to the SDS-PAGE separating gel (12.5-17.5% acrylamide). For Western blots, the gel was soaked in transfer buffer with 1 mM EDTA to remove the Mn 2+ ions before transfer.
Structure solution and refinement. The starting phases for the structures were obtained using molecular replacement with the PRL2-CBSpair complex (PDB entry 5K22) in Phaser (58). The resulting models were extended manually with the help of the program Coot (59) and improved by several cycles of refinement using PHENIX (60). The final models have good stereochemistry (Suppl. Tumor formation assay. Tumor formation experiments were performed as described previously (4). Briefly, one million B16 cells stably expressing Myc-PRL3 were injected into the tail veins of female C57BL/6J mice (Charles River Laboratories). Mice were euthanized 3 weeks after injection, the lungs were excised, and the black spherical B16 colonies on their surface larger than 0.1 mm in diameter were counted.
Mg 2+ depletion experiments. HeLa cells were cultured in DMEM supplemented with 10% FBS and antibiotics. For Mg 2+ depletion experiment, cells were cultured in Mg 2+ -free DMEM (Wisent) and dialyzed FBS (Wisent) with or without 1 mM MgSO4. Cells were lysed in RIPA buffer (50 mM HEPES pH 7.5, 150 mM NaCl, 1%NP-40, 1% Na deoxycholate, 0.1% SDS) and 5 μg total protein loaded per well on SDS-PAGE gels with and without the Phos-tag reagent. Proteins transferred to a PVDF membrane. The membrane was blocked and first incubated with mouse anti-PRL2 (Millipore), then with HRP-conjugated anti-mouse secondary antibody (Jackson ImmunoResearch 115-035-062 1:5,000). Chemiluminescent detection on film was done using the Amersham ECL Prime detection kit (GE Healthcare).
Cysteine phosphorylation in tissues. Frozen mouse tissues (gift of Philippe Gros) were homogenized and lysed in RIPA buffer. Total protein was estimated using Pierce BCA Protein assay kit and the amount for SDS-PAGE analysis adjusted to give roughly equal amounts of PRL staining: colon (10 μg / lane), heart (40 µg), liver (40 µg), and skeletal muscle (20 µg). Where indicated, samples were boiled for 10 min in loading buffer prior to electrophoresis. Western blotting was done as indicated for HeLa cell extracts.

Data Availability
The atomic coordinates and structure factors (codes 6WUR and 6WUS) have been deposited in the Protein Data Bank (http://wwpdb.org/).

Acknowledgements
We thank Nassima Fodil (Gros laboratory) and Sebastien Tabaries (Siegel laboratory) for assistance with the mouse tissues. Crystallographic data were acquired at the Canadian Light Source (CLS) and at the Macromolecular Diffraction (MacCHESS) facility at the Cornell High Energy Synchrotron Source (CHESS). CHESS is supported by National Science Foundation award DMR-0225180 and National Institutes of Health/NCRR Grant RR-01646.

Funding
This work was supported by Natural Sciences and Engineering Research Council of Canada grant RGPIN-2020-07195 to K. G. Fitting parameters shown in Suppl. Fig. S1.