PRMT5 C-terminal Phosphorylation Modulates a 14-3-3/PDZ Interaction Switch*

PRMT5 is the primary enzyme responsible for the deposition of the symmetric dimethylarginine in mammalian cells. In an effort to understand how PRMT5 is regulated, we identified a threonine phosphorylation site within a C-terminal tail motif, which is targeted by the Akt/serum- and glucocorticoid-inducible kinases. While investigating the function of this posttranslational modification, we serendipitously discovered that its free C-terminal tail binds PDZ domains (when unphosphorylated) and 14-3-3 proteins (when phosphorylated). In essence, a phosphorylation event within the last few residues of the C-terminal tail generates a posttranslational modification-dependent PDZ/14-3-3 interaction “switch.” The C-terminal motif of PRMT5 is required for plasma membrane association, and loss of this switching capacity is not compatible with life. This signaling phenomenon was recently reported for the HPV E6 oncoprotein but has not yet been observed for mammalian proteins. To investigate the prevalence of PDZ/14-3-3 switching in signal transduction, we built a protein domain microarray that harbors PDZ domains and 14-3-3 proteins. We have used this microarray to interrogate the C-terminal tails of a small group of candidate proteins and identified ERBB4, PGHS2, and IRK1 (as well as E6 and PRMT5) as conforming to this signaling mode, suggesting that PDZ/14-3-3 switching may be a broad biological paradigm.

Arginine methylation is a common PTM 3 that alters roughly 0.5% of all arginine residues in the cells. There are three types of arginine methylation: monomethylarginine, asymmetric dimethylarginine, and symmetric dimethylarginine (1). PRMT5 is one of nine PRMTs, and it is responsible for the vast majority (Ͼ95%) of the symmetric dimethylarginine modifications (2). PRMT5 was first characterized as a transcriptional repressor for cyclin E1 (3), and in this context, it methylates histone H3R8me2s, H2AR3me2s, and H4R3me2s (4). An epigenetic silencing role for PRMT5 has also recently been reported for the cell cycle inhibitor p21 (5).
However, PRMT5 clearly has a number of non-histone substrates that are localized to the cytoplasm and the plasma membrane (6). In the cytoplasm, PRMT5 forms part of the methylosome and methylates a number of splicing factors (7). In keeping with these observations, the conditional deletion of PRMT5 in neural stem cells leads to defects in the core splicing machinery, reduced constitutive splicing, and massive alterations in alternative splicing profiles (8). Thus, this arginine methyltransferase has key biological roles that are associated with each of the major cellular compartments (the nucleus, the cytoplasm, and the plasma membrane), although little is known about how the activity and localization of PRMT5 in these different compartments are regulated.
There is an emerging interest in establishing how signal transduction pathways communicate with chromatin and regulate changes to the epigenetic landscapes (9). It is likely that enzymes like PRMT5 may be marked by different PTMs to alter their activity and subcellular localization. Most notably, tyrosine phosphorylation of PRMT5 by mutant Jak2 prevents its interaction with a critical cofactor MEP50, thereby inhibiting methylation of histone substrates (10). Here we have identified a signaling module at the C-terminal tail of PRMT5 that can regulate its subcellular localization through threonine phosphorylation and selectively prevents PDZ interactions and facilitates 14-3-3 binding.
PDZ domains are one of the most abundant protein domains found in multicellular eukaryotes (11). Over 100 human proteins harbor PDZ domains, often in multiple copies per protein (there are roughly 250 different human PDZ domains) (12). PDZ domains generally bind short C-terminal motifs in their ligands, with the last amino acid being a hydrophobic residue (13). Large scale screening approaches have classified the different motifs that can be bound by PDZ domains (14,15). The plasticity of PDZ interactions can be tuned by the phosphorylation of Ser/Thr/Tyr residues found within the PDZ bind-ing motifs (16,17). The multidomain nature of many PDZ domain-containing proteins implicates them in biological processes that involve scaffolding functions like 1) the clustering of ion channels and signaling receptors on membranes, 2) the cross-talk between the plasma membrane and cytoskeletal structures, and 3) the maintenance of cell polarity.
14-3-3s were the first proteins to be identified as "readers" of phospho-serine/threonine motifs (18). There are seven different, highly related, 14-3-3 isoforms. They can assemble as stable homo-and heterodimers, which is critical for many of their biological functions. These functions include 1) blocking protein-protein interactions, 2) trapping proteins in the cytoplasm that normally shuttle into the nucleus, 3) regulating enzyme activity, 4) bridging between enzymes and substrates, and 5) protecting their binding partners from protein degradation pathways (19). An unbiased selection of 14-3-3 protein-binding peptides and a comparison of identified binding proteins has revealed three types of consensus sequences capable of mediating phospho-dependent interactions with 14-3-3s (20,21). Interestingly, one of these motifs (motif III) is located at the extreme C terminus of several proteins and displays remarkable overlap with a PDZ binding motif (22).
The human papillomavirus E6 oncoprotein harbors the ability to bind PDZ domains though a PDZ binding motif (23,24). Embedded within the PDZ binding motif of E6 is an Akt/PKA phosphorylation site, which when modified inhibits PDZ domain binding (25). Recently, it was discovered that this C-terminal phosphorylation event not only blocks PDZ interactions, but also generates a docking motif for 14-3-3 proteins (26), thus establishing the first example of PDZ/14-3-3 switching in mammalian cells, albeit not with a mammalian protein.
Our study identifies a similar switching motif in PRMT5 and reveals that this switch may be broadly used in biological signaling.

Results
PRMT5 Is Phosphorylated at Its C Terminus-Most arginine methyltransferases display robust intrinsic enzymatic activity in in vitro methylation assays (27), suggesting that in a cellular context, this enzymatic activity must be regulated, possibly by PTM of the PRMT enzymes themselves. To test this hypothesis, we scanned the amino acid sequences of the nine PRMTs using Web-based software that predicted the kinase-specific phosphorylation sites (Scansite and NetPhorest) (28,29). Interestingly, PRMT5 was predicted, with high stringency, to harbor an Akt phosphorylation site at residue Thr-634, at the Ϫ4 position of the C-terminal end of the enzyme (Fig. 1A). To experimentally determine whether this was indeed an Akt phosphorylation site, we fused the second half of PRMT5 to GFP and mutated the threonine residue of interest to an alanine (T634A), by site-directed mutagenesis. GFP-PRMT5 and GFP-PRMT5 T634A expression vectors were then co-transfected into FIGURE 1. PRMT5 is phosphorylated at threonine 634 in its C terminus. A, schematic representation of GFP-PRMT5 (amino acids 340 -637). The predicted Akt-phosphorylated site (Thr-634) is shown. The signature methyltransferase motifs are boxed. B, HeLa cells were transfected with GFP-PRMT5 (WT or T634A) and Myr-Akt and then treated with 0.1 M calyculin A. Cells were lysed and subjected to immunoprecipitation (IP) with ␣-GFP. Immunoprecipitates and input samples were then blotted with pan-phosphothreonine (␣-pT) and ␣-GFP antibodies. C, SK-CO15 cells transiently expressing GFP-PRMT5 (WT or T634A) were treated with angiotensin II, dexamethasone, and forskolin or co-transfected with Myr-Akt. Cell lysates were immunoprecipitated with ␣-GFP followed by Western blotting with pan-phosphothreonine and ␣-GFP antibodies. The input samples were blotted with ␣-Akt. ␤-Actin was used as a loading control. D, the bar graph represents the normalized ratio of phosphorylated GFP-PRMT5 (pan-phosphothreonine) and total GFP-PRMT5 (␣-GFP). Quantification was performed by densitometry of the top panel in C and a duplicate experiment. Three independent measurements were taken for each band from the two different blots. S.D. is denoted by error bars; *, p Ͻ 0.01; **, p Ͻ 0.001, generated by an unpaired Student's t test.
HeLa cells expressing a constitutively active Akt (Myr-Akt), and phosphorylation was detected using a pan-phosphothreonine antibody. Phosphorylation of wild-type but not mutant PRMT5 was observed (Fig. 1B).
Next, we used an unbiased screening approach to identify kinases that can phosphorylate this site on PRMT5. This large scale in vitro kinase assay was performed by Kinexus using a C-terminal PRMT5 peptide, over 295 different recombinant kinases, and radiolabeled ATP ( Fig. 2A). The peptide used in the in vitro phosphorylation screen contained two Thr residues, as well as a Ser and a Tyr residue (Fig. 2B). To confirm that the primary phosphorylation site on this peptide is Thr-634, we performed a secondary screen that compared the phosphorylation efficiency of the top 18 kinases identified in the primary screen, on PRMT5 T634 and PRMT5 A634 peptides (Fig. 2B). This screen confirms that the C-terminal threonine of PRMT5 can be phosphorylated in vitro by Akt and also by calcium/calmodulin-dependent protein kinase II, serum-and glucocorticoidinducible kinases (SGKs), and protein kinase A (PKA) family members. To establish whether these different kinases could phosphorylate PRMT5 in cells, cells were treated with angiotensin II (calcium/calmodulin-dependent protein kinase II activation) (30), dexamethasone (SGK activation) (31), and forskolin (PKA activation) (32), and with myristoylatable Akt as a positive control. Both Akt and SGK kinases significantly increased phosphorylation of the Thr-634 site, suggesting that these are the relevant kinase families that can signal to PRMT5 (Fig. 1, C and D).
The Phosphorylated C Terminus of PRMT5 Binds 14-3-3s-Similarly to PRMT5, Akt and SGK phosphorylate the FoxO family of transcription factors on common sites (33, 34), gener-ating 14-3-3 docking motifs (35). Indeed, Akt has a consensus phosphorylation site that closely resembles the recognition motifs for 14-3-3 binding, and many Akt substrates are 14-3-3 ligands (21). A NetPhorest scan of PRMT5, which predicted the Akt phosphorylation site at its C terminus, also predicted that this phosphomotif was a 14-3-3 binding site. 14-3-3 proteins have three different binding motifs, and one of these motifs (motif III) is located at the extreme C terminus of several 14-3-3 ligands (Fig. 3A). To test the hypothesis that the phosphorylated C-terminal tail of PRMT5 interacts with 14-3-3 proteins, we synthesized a biotinylated PRMT5 peptide set that was unmodified (PRMT5(623-637)) or phosphorylated (PRMT5 pT634 ) (Fig. 3B). These peptides were then preconjugated to streptavidin-Cy3 and -Cy5 and used to probe a protein domain microarray (36) on which all seven 14-3-3 proteins were represented. We found the PRMT5 phosphopeptide bound the 14-3-3s, whereas the unphosphorylated PRMT5 control peptide did not (data not shown). Unexpectedly, the control peptide (but not the phosphopeptide) bound a few PDZ domains that were also FIGURE 2. In vitro screening of kinases that phosphorylate a PRMT5 C-terminal peptide. A, the activity of 295 protein kinases was tested against a PRMT5 C-terminal peptide. These assays were based on the direct quantification of radiolabeled phosphate from ATP (␥-33 P) on to the peptide substrate. These assays were performed by Kinexus. In vitro kinase activity was ranked as excellent (Ͼ5.5 pmol/min), good (4.6 -5.4 pmol/min), low (1.8 -4.5 pmol/ min), and non-phosphorylated (Ͻ1.7 pmol/min). B, the 18 best kinases (excellent and good ranking) were used to validate the initial in vitro phosphorylation screen using PRMT5 WT and T634A C-terminal peptides as substrates. arrayed on the slide (data not shown). Analysis of the PRMT5 sequence revealed a hydrophobic residue at the extreme C-terminal end of the protein (Fig. 3A), which closely matches one of the predicted binding motifs (motif I) for PDZ domains. These data suggested that the C-terminal end of PRMT5 was a potential regulatory hub, where phosphorylation, 14-3-3 binding, and PDZ binding all intersected.
The Unphosphorylated C Terminus of PRMT5 Binds PDZs-To investigate this potential node of signaling in detail, we cloned and produced a library of mouse GST-PDZ domains and generated a focused PDZ/14-3-3 protein domain microarray. There are roughly 250 PDZ domains in the mouse proteome. A detailed analysis of 157 recombinant and soluble PDZ domains, performed by the MacBeath group (15), revealed that a subset of 87 domains bound at least one ligand within a test set of 217 different peptides. We subcloned and expressed this set of 87 validated PDZ domains, 11 of which displayed solubility issues. Thus, the focused PDZ/14-3-3 microarray harbors 76 different mouse PDZ and all seven 14-3-3 proteins (Fig. 4). We probed this array with the PRMT5 peptide set, unmodified PRMT5(623-637) or phosphorylated PRMT5 pT634 , and confirmed that the phosphopeptide bound 14-3-3 domains, whereas the unphosphorylated peptide bound several PDZ domains (Fig. 5, A and B). As a positive control, we tested the E6 oncoprotein C-terminal peptide set that was recently reported to undergo PDZ/14-3-3 switching in a phospho-dependent manner (26).
We also performed a pilot in silico screen using Scansite and NetPhorest to identify proteins with predicted overlapping Akt phosphorylation sites and 14-3-3 and PDZ binding sites. We identified three such proteins: 1) the receptor tyrosine kinase ERBB4, 2) the inward rectifier potassium channel IRK1 (Kir2.1), and 3) the prostaglandin-endoperoxide synthase PTGS2 (COX2) (Fig. 3B). We tested the phosphorylated and unphosphorylated peptide sets from these three candidate proteins and observed that all three undergo PDZ/14-3-3 switching, with phosphopeptides binding 14-3-3s and unphosphorylated peptides binding PDZ domains (Fig. 5, A and B). This finding expands the number of proteins containing a PDZ/14-3-3 "switch" beyond PRMT5 and the E6 viral protein to other mammalian proteins that play key roles in the cell and opens the door to identifying the function of phospho-dependent PDZ/14-3-3 switching in several different biological processes.
Endogenous PRMT5 Interacts with Recombinant 14-3-3s and the NHERF2 PDZ Domain-Microarray data predicted that the unmodified PRMT5(623-637) peptide interacts strongly with the PDZ domains of NHERF2, MPP7, and GRIP1 and weakly with NHERF1, PDZ-LIM5, and SCRIB PDZ domains (red dots in Fig. 5A). Interestingly, the PDZ domain of PDZ-LIM2 is unique, because it binds both the phosphorylated and unphosphorylated tail of PRMT5 (yellow dots in Fig. 5A). To independently validate the PDZ/PRMT5 interaction detected by the microarray approach, we performed peptide pull-downs of the seven GST-PDZs and GST alone as a negative control (Fig. 6A). This allows a rough comparison of the relative strength of the different interactions, because the GST input can be well controlled (Fig. 6A, bottom). Using this approach, we found that the unphosphorylated tail of PRMT5 interacted most strongly with the GST fusion of full-length NHERF2, which harbors two PDZ domains.
Next, we investigated which of the two NHERF2 PDZ domains interacted with the C-terminal peptide of PRMT5 and also confirmed the PDZ/14-3-3 phospho-switch by peptide pull-down. The four indicated GST fusion proteins (Fig. 6B) were subjected to pull-downs with biotinylated PRMT5(623-637) and PRMT5 pT634 peptides. The first PDZ domain of NHERF2 interacts with the PRMT5(623-637) peptide only when it is unphosphorylated, whereas the PRMT5 pT634 peptide interacts with 14-3-3⑀ (Fig. 6C). This finding confirms the phospho-switch we first observed on the array (Fig. 5A). To determine whether NHERF2 and 14-3-3⑀ interacted with endogenous PRMT5 in a similar way, we performed GST pulldowns with NHERF2 and 14-3-3⑀ from lysates of HeLa cells cultured in the presence or absence of the phosphatase inhibitor, calyculin A (CalA). Under hypophosphorylated conditions, endogenous PRMT5 interacts with NHERF2, and in the presence of CalA, PRMT5 interacts with the 14-3-3 protein (Fig. 7A).
In the context of the full-length endogenous PRMT5 protein, it is possible that the pull-downs we observe with NHERF2 and/or 14-3-3⑀ are not due to interactions with the C terminus. We thus performed a similar GST pull-down experiment but this time used full-length Myc-tagged PRMT5 and two different C-terminal mutants of the PRMT5 expression vector, one with a deletion of the last 6 amino acids of PRMT5 (⌬) and the other with a single amino acid change (T634A). Importantly, PRMT5 homodimerizes (6), and to reduce the confounding influence of the tagged PRMT5 expression vectors interacting with endogenous PRMT5, we performed these experiments in PRMT5 shRNA knockdown HeLa cells, using shRNA-resistant expression vectors. Again, we see that CalA treatment is required for the 14-3-3 interaction and inhibits the PDZ inter- FIGURE 5. Phosphorylation triggers switching between 14-3-3 and PDZ interactions. A, PRMT5, ERBB4, E6 (HPV16), PGHS2, and IRK1 unphosphorylated and phosphorylated peptides were labeled with Cy5 (red) and Cy3 (green), respectively, and used to probe a protein microarray containing PDZ domains and 14-3-3 GST fusion proteins. The bottom right panel shows the array probed with ␣-GST for the loading control. The PDZ domains (red) and 14-3-3 proteins (green) are blocked. B, graphical depiction of the interactions observed in A. Red and green squares, interactions with unphosphorylated and phosphorylated peptides, respectively. FEBRUARY 10, 2017 • VOLUME 292 • NUMBER 6 action. Significantly, either truncation of PRMT5 or the introduction of a single amino acid change results in total loss of both PDZ and 14-3-3 binding (Fig. 7B). In addition, weak co-immunoprecipitation of ectopically expressed Myc-tagged PRMT5 and GFP-tagged 14-3-3 was observed, and this interaction required an intact Thr-634 phosphorylation site (Fig. 6D).

14-3-3/PDZ Switching
The C Terminus of PRMT5 Localizes It to the Plasma Membrane-Most PDZ domain-containing proteins are membrane-associated, they frequently have scaffolding functions, and their ligands are commonly transmembrane receptors and ion channels (12). Indeed, of the seven PDZ domains identified as interacting with the C-terminal tail of PRMT5 (Fig. 5, A and  B), six are associated with the plasma membrane, including the strongest binder, NHERF2. Importantly, PRMT5 has been shown to methylate a number of proteins that are membraneassociated (37,38). We thus surmised that the C terminus of PRMT5 was responsible for localizing it to the intracellular side of the plasma membrane.
To test this hypothesis, we chose to work in SK-CO15 cells, which are human intestinal epithelial cells that express high levels of NHERF2 (39). First, we performed cell fractionation studies to determine whether endogenous PRMT5 is localized to the membrane. Indeed, we found that PRMT5 was both cytoplasmic and membrane-associated (Fig. 7C). 14-3-3 proteins predominantly associated with the cytoplasmic fraction, and NHERF2 predominantly associated with the membrane fraction. We next asked whether the C terminus of PRMT5 was required for its recruitment to the plasma membrane. Fulllength Myc-tagged PRMT5 and the C-terminal deletion mutant (PRMT5 ⌬ ) were expressed in SK-CO15 cells, and after cell fractionation, abundant full-length Myc-PRMT5 was found in the membrane fraction, whereas Myc-PRMT5 ⌬ was dramatically reduced (Fig. 7D, top right). The small amount of Myc-PRMT5 ⌬ that is present in the membrane fraction is probably due to homodimerization with endogenous PRMT5 and represents indirect membrane retention. The 14-3-3/PDZ-interacting Motif of PRMT5 Is Required for Mouse Viability-To evaluate the importance of the 14-3-3/ PDZ switch in vivo, we used CRISPR/Cas9 to generate a mouse in which we replaced the last 6 amino acids of PRMT5 with an HA tag (PRMT5 ⌬HA ). Two founder PRMT5 ⌬HA knock-in mice were obtained (lines 2 and 6) that had the expected insertion of

14-3-3/PDZ Switching
an HA tag and the concomitant removal of the last 6 residues of PRMT5 (Fig. 8A). Two additional founders (lines 8 and 10) displayed alterations at the C-terminal tail due to indel-mediated frameshifts (Fig. 8A). Lines 6 and 10 were bred, and no homozygous embryos (of 71 embryos examined) were detected at embryonic day 9.5 and 11.5, respectively, thus establishing that the C-terminal end of PRMT5 is required for mouse viability, whether the tail is replaced with an HA tag or is altered to a foreign sequence due to an indel.
Importantly, analysis of embryo lysates from the PRMT5 ⌬HA intercross shows that HA tagged PRMT5 is expressed in the heterozygous embryos and that PRMT5 levels are normal (not destabilized by the presence of the tag) (Fig. 8B). Thus, the embryonic lethality is probably due to mislocalization of the enzyme (because it cannot interact with PDZs and 14-3-3s) and not due to destabilization of the enzyme. Heterozygous embryos display no obvious phenotype at embryonic day 11.5 (Fig. 8C). We next attempted to develop homozygous PRMT5 ⌬HA ES cell lines from blastocysts derived from intercrosses of mice heterozygous for PRMT5 ⌬HA . We were able to establish wild-type (4 clones) and heterozygous (6 clones) lines but not homozygous lines, which strongly suggests that the switching motif is required for the derivation of ES cells, as has been reported for PRMT5 null mice (40,41).

Discussion
Possible Roles for PRMT5 on the Plasma Membrane-We found that PRMT5 associates with the plasma membrane and that this association is dependent on its C-terminal PDZ binding motif (Fig. 7). It is likely that the concentration of PRMT5 at the cell membrane promotes the methylation of PRMT5 sub-strates that are transmembrane or membrane-associated proteins. Indeed, PRMT5 has been shown to methylate a host of plasma membrane-associated proteins, including the EGF receptor (38), the D2-like dopamine receptor (42), sodium channels (37), and srGAP2, which has been implicated in lamellipodia and filopodia formation (43), and to associate with the TRAIL receptor (44).
Possible Roles for the PRMT5/14-3-3 Interactions-PRMT5 has been identified as a 14-3-3-interacting protein in two independent large scale protein-protein interaction screens. It was first found as a 14-3-3 target by the Pawson group using an LC-MS/MS approach (45) and more recently identified in a high density interactome screen, which used a yeast two-hybrid approach (46). There are a number of possible functions for the phospho-dependent PRMT5/14-3-3 interaction. First, 14-3-3s could sequester PRMT5 and prevent its trafficking into the nucleus and its roles in epigenetic signaling. Second, the 14-3-3 interaction could prevent the dephosphorylation of PRMT5 and thus its re-engagement with PDZ domain-containing proteins on the plasma membrane. Third, because 14-3-3s homo-/ heterodimerize, the interaction may stimulate and stabilize the homodimerization of PRMT5 itself, thereby regulating PRMT5 activity. Fourth, 14-3-3 interactions may form a bridge between PRMT5 and its substrates. Thus, C-terminal phosphorylation of PRMT5 could promote the methylation of a unique subset of cytoplasmic substrates. Last, it is possible that the 14-3-3s (and any of the six PDZs that interact with PRMT5) are direct substrates for this enzyme. Our inability to generate somatic or ES cell lines that only express the PRMT5 ⌬HA has hampered our efforts to explore these possibilities. . PRMT5 interacts with NHERF2 and associates with the plasma membrane. A, control and PRMT5 knockdown (KD) HeLa cells were treated with or without calyculin A, and the lysates were used for GST pull-down assays. 14-3-3⑀ was able to pull down PRMT5 in the calyculin A-treated cells, whereas NHERF2 pulled down PRMT5 from the untreated cell lysate. B, PRMT5 KD HeLa cells were transiently transfected with shRNA-resistant Myc-PRMT5 WT, ⌬ (deletion of last 6 amino acids), and T634A mutant. Cells were subjected to calyculin A treatment, and pull-down assays were performed as described in A. C, SK-CO15 cells, which express high levels of NHERF2, were fractionated by ultracentrifugation. Fractions were immunoblotted with antibodies against PRMT5, NHERF2, cadherin family members (a membrane marker), and 14-3-3 family members. D, SK-CO15 cells were transiently transfected with Myc-tagged PRMT5 and PRMT5 ⌬ . Cells were fractionated and analyzed as in C. Fractions were tested with the indicated antibodies, and localization of Myc-tagged PRMT5 constructs was evaluated.
How Broad Is the PDZ/14-3-3 Switching Paradigm?-Virus activity in mammalian cells often co-opts normal cellular processes and thus can provide us with insights into critical signaling/regulatory pathways that are active in mammalian cells (47). Prime examples of this are the HPV E6 and E7 oncoproteins, which target p53 for degradation and sequester phosphorylated pRb, respectively (48,49). The E6 oncoprotein not only targets p53 but also has other cellular activities, including the ability to bind PDZ domains though a class I PDZ binding motif (23,24). Phosphorylation of the PDZ binding motif of E6 blocks PDZ domain binding (25) and facilitates 14-3-3 protein binding (26). Utilization of a PDZ/14-3-3 "switch" by E6 infers the presence of a similar "switch" in mammalian cells. Indeed, here we show that PRMT5 undergoes a phospho-dependent switch between PDZ and 14-3-3 binding modes, and furthermore, a small screen of additional candidates identified ERBB4, PGHS2, and IRK1 that adhere to this switching paradigm. To determine the breadth of this paradigm, it will be important to expand this in silico screen, in conjunction with experimen-tal validation on PDZ/14-3-3 protein domain microarrays. It is very likely that there will be many more mammalian proteins that undergo this type of switching.

Plasmid Constructs
Mammalian Expression Vectors-All expression constructs containing PRMT5 are derivatives of human wild-type cDNA. The Myc-PRMT5 construct was cloned into the pVAK vector containing a N-terminal Myc epitope tag (EQKLISEEDL). EcoRI sites flank PRMT5 sequences. PRMT5 Myc-PRMT5 T634A and Myc-PRMT5 ⌬ were generated by mutagenesis, using the QuikChange site-directed mutagenesis kit according to the manufacturer's protocol (Agilent). The desired mutation was verified by DNA sequencing. GFP-PRMT5 encoding PRMT5 amino acids 340 -637 was cloned into pEGFP-C1 vector (Clontech) using the EcoRI site. GFP-PRMT5 T634A was obtained by site-directed mutagenesis as described above. HeLa PRMT5 stable knockdown and control cell lines were obtained from Dr. Dent (50).
Bacteria Expression Vector-The PDZ domain array content was selected considering the biophysical interactions described by Stiffler et al. (15). The protein content is listed in Fig. 4. Human 14-3-3 proteins encoding the full-length sequence and mouse PDZ domains are codon-optimized for bacterial expression and synthesized by Biomatik. Constructs are flanked by BamHI and XhoI restriction sites into pGEX6P-1 plasmid. Human NHERF2 constructs were cloned by PCR into pGEX6p-1 using BamHI and EcoRI; plasmids encode the fulllength protein, NHERF2-PDZ 1 (amino acids 1-152), and NHERF2-PDZ 2 (amino acids 107-337). The 14-3-3⑀ insert was subcloned into pEGFPC1.

Recombinant Protein Purification
GST fusion proteins were purified following a standard method. Briefly, protein was expressed in BL21 cells for 4 h at 37°C with 0.1 mM isopropyl-␤-D-thiogalactopyranoside. Cells were resuspended in PBS buffer and then lysed by sonication at 30% amplitude for 10 s. Lysates were cleared by centrifugation and incubated with glutathione-Sepharose 4B resin at 4°C with tumbling (GE Healthcare). Subsequently, Sepharose-immobilized GST-tagged proteins were washed with PBS and eluted with elution buffer (100 mM Tris-HCl, pH 8.0, 120 mM NaCl, and 40 mM reduced glutathione).

GST Pull-down Assay
GST-tagged proteins (ϳ10 g) were incubated with cell lysates (from one confluent 10-cm dish, lysed on 1 ml of mild buffer (50 mM Tris-HCl, pH 7.5, 150 mM NaCl, 0.1% Nonidet P-40, 5 mM EDTA, 5 mM EGTA, 15 mM MgCl 2 , and proteinase inhibitor mixture (Roche Applied Science))) for 2 h at 4°C with tumbling at a final volume of 1 ml. Equilibrated glutathione-Sepharose 4B resin was then added for an additional 1 h with tumbling. Samples were then washed four times with mild lysis buffer before elution, SDS-PAGE, and Western blotting. Mouse lines 2 and 6 display the expected replacement of the last 6 amino acids with an HA tag. Mouse lines 8 and 10 represent alterations at the C terminus tail due to indels, resulting in a shift of reading frame. B, Western blotting analysis of embryonic day 11.5 embryo lysates using ␣-HA and ␣-PRMT5 antibodies. ␤-Actin was used as a loading control. C, the gross phenotype of WT and heterozygous PRMT5 ⌬HA embryos at embryonic day 11.5.

14-3-3/PDZ Switching Transient Transfection of Cultured Cells
HeLa and SK-CO15 cell lines were cultured in high glucose DMEM (Sigma) supplemented with penicillin/streptomycin (Gibco), MEM non-essential amino acids (Gibco), 15 mM HEPES, and 10% heat-inactivated fetal bovine serum (FBS), under standard conditions of 37°C and 5% CO 2 . Transfection of cells with mammalian expression constructs by Lipofectamine 2000 (Invitrogen) was done according to the methods provided by the manufacturer's specifications. Cells were transfected overnight and harvested the following day for analysis.

Cell-based Phosphorylation Assay
SK-CO15 cells were transiently transfected with GFP-PRMT5 and GFP-PRMT5 T634A and treated with 30 M forskolin (CST) for 30 min, 1 nM angiotensin II (Sigma) for 45 min, 1 M dexamethasone (Sigma) for 24 h, followed by treatment with 0.05 M calyculin A for 10 min. Immunoprecipitated samples were blotted with a pan-anti-phosphothreonine antibody. Quantification was done using ImageJ software, measuring the band intensities of immunoprecipitated samples as detected by a pan-␣-Thr(P)-specific antibody and ␣-GFP. Experiments were performed in duplicate with similar results.

In Vitro Kinase Assay
The PRMT5 C terminus wild type (WT) peptide (KKPTGR-SYTIGL-COOH) was used for these assays. In vitro phosphorylation assay were performed on the PRMT5 WT peptide by Kinexus, using a radiometric assay [␥-33 P]ATP and 295 different recombinant kinases. In a second stage, kinases from the first stage that generated signals Ͼ250 cpm (4.6 pmol/min) were retested against the PRMT5 WT peptide and the PRMT5 T634A mutant peptide (KKPTGRSYAIGL-COOH). The in vitro kinase assay was performed at ambient temperature for 20 -40 min in a final volume of 25 l containing a mixture of 10 -50 nM active protein kinase, peptide, and 5 l of [␥-33 P]ATP (250 M stock solution, 0.8 Ci) in a suitable buffer. After the incubation period, 10 l of the reaction mixture was spotted onto a phosphocellulose P81 plate and counted in the presence of scintillation fluid in a Trilux scintillation counter. The mass transfer was calculated by the equation, mass transferred ϭ (cpm peptide ϫ 1250 pmol)/(cpm ATP ϫ 30 min); specific radioactivity of 5 l of ATP is 2,267,128 cpm.

Antibodies
The following antibodies were used in this study: ␣-PRMT5

GFP Immunoprecipitation
For GFP immunoprecipitations, ␣-GFP magnetic Sepharose and agarose resin (Allele Biotech) was used. For Myc immunoprecipitations, ␣-Myc antibody and Dynabeads Protein A (Life Technologies) were used. Briefly, cells were harvested in PBS buffer and then pelleted by centrifugation. The cell pellet was resuspended in 200 l of lysis buffer (20 mM Tris-HCl, pH 7.5, 150 mM NaCl, 1 mM EDTA, 0.5% Nonidet P-40, proteinase inhibitor mixture, and phosphatase inhibitors) and then lysed by sonication (10 cycles of 30 s on/off using Bioruptor, Diagenode). After centrifugation at 20,000 ϫ g for 10 min at 4°C, the supernatant was collected and diluted with binding buffer (10 mM Tris-HCl, pH 7.5, 150 mM NaCl, proteinase inhibitor mixture, and phosphatase inhibitors) to a final volume of 700 l. Antibodies were added to the lysates and incubated at 4°C overnight. Equilibrated beads were incubated with lysate with tumbling for 1 h at 4°C. The immunoprecipitated samples were washed in buffer (10 mM Tris-HCl, pH 7.5, 500 mM NaCl, proteinase inhibitor mixture, and phosphatase inhibitors) using a magnetic stand. The beads were then resuspended in 30 l of loading buffer and boiled in loading buffer for 5 min to elute the proteins. The samples were subjected to Western blotting analysis.

Western Blotting Analysis
Samples were separated by SDS-PAGE. Proteins were transferred onto PVDF membrane using a semi-wet transfer apparatus. Membranes were blocked in blocking buffer (PBS, 0.1% Tween 20, and 5% milk) for 1 h at room temperature and then incubated with primary antibody in the blocking buffer overnight at 4°C. The blots were then washed, probed with an HRPconjugated secondary antibody, and detected using ECL reagents (Amersham Biosciences).

Peptide Pull-down
15 g of biotin-labeled peptides were immobilized on streptavidin-agarose beads (Sigma) in peptide-binding buffer (50 mM Tris-HCl, pH 7.5, 150 mM NaCl, 1 mM EDTA, 2 mM dithiothreitol, 0.5% Nonidet P-40) overnight at 4°C and washed three times with binding buffer to remove unbound peptides. To reduce nonspecific binding, GST-fused proteins were preincubated with streptavidin-agarose beads overnight. Samples were centrifuged, and the supernatants were incubated with immobilized peptide in 300 l of binding buffer overnight. The beads were washed three times using binding buffer and then resuspended in 30 l of loading buffer and boiled for 5 min. The samples were analyzed by SDS-PAGE and immunoblotted with ␣-GST antibody.

Protein Microarray
Protein arrays were generated as described previously (36). Briefly, proteins were arrayed in duplicate, at different angles to facilitate rapid identification, using an Aushon 2470 microarrayer (Aushon Biosystems). GST fusion proteins were arrayed from a 384-well plate, which contained 10 l of each protein at a approximate concentration of 1 g/l. The protein array was composed of 10 blocks (A-J), each in a 4-row by 5-column format, with a distance of 600 m between spots. GST alone was printed at the lower right side of each block. Proteins were printed on nitrocellulose-coated glass slides (Grace Bio-labs). To generate the probe, 10 g of biotinylated peptide was bound to 5 g of Cy3-streptavidin or Cy5-streptavidin (GE Healthcare) in 500 l of PBST (PBS ϩ 0.1% Tween 20). Labeled peptides were cleared of unconjugated streptavidin label by incubation with biotin-agarose beads (Sigma). Arrays were probed with fluorescently labeled peptides overnight at 4°C, and unbound peptides were washed away with PBST. The fluorescent signal was detected using a GenePix 4200A microarray scanner (Molecular Devices). 550-and 675-nm filters were used for the detection of Cy3-and Cy5-labeled probes. GST signal was detected with ␣-GST and 555-conjugated rabbit secondary antibodies.

Membrane Fractionation
This protocol was adapted from Ref. 51. Three 15-cm plates harboring cells at 80% confluence were used. Cells were washed three times with homogenized buffer (25 mM imidazole, 250 mM sucrose, 1 mM EDTA, pH 7.2, and proteinase inhibitor) and harvested by scraping into 10 ml of homogenizing buffer. The cells were lysed with six passes through an 18-gauge needle and then six times through a 27-gauge needle. Lysate was centrifuged at 5,800 ϫ g for 15 min, and the supernatant was collected and then centrifuged again at 47,800 ϫ g for 30 min. The resulting supernatant contained the cytoplasmic proteins, and the pellet contained the membrane and membrane-associated proteins. Pellet was resuspended in 1 ml of homogenizer buffer containing proteinase inhibitor. The samples were analyzed by SDS-PAGE and immunoblotting.

Generation of the PRMT5 ⌬HA Mouse Model and Genotyping
CRISPR/Cas9 was used to generate a mouse model in which we replaced the last 6 amino acids of PRMT5 with an HA tag, which we refer to as PRMT5 ⌬HA . To generate the PRMT5 ⌬HA knock-in mice, we co-injected sgRNA, Cas9 mRNA, and an oligonucleotide donor into one-cell stage mouse embryos. The oligonucleotide donor is a double-stranded gBlock fragment that has 80 bp of homology in each arm, with DNA encoding the HA tag (60 bp) in the middle: TGCAGCAATTCCAAG-AAAGTGTGGTACGAGTGGGCGGTGACGGCCCCCGT-CTGTTCTTCTATTCACAACCCTACCGGCCGGGGAT-ATCCATATGATGTTCCTGATTATGCTTAGCCCTGC-ACACAGTGTCAAAACCTTGGAAGCAGCTCTGAGTT-CTCTTCCTACAGCACAGAAGGTGTAGAACA. The gRNA/ Pam used in this model is ATGGTATAGGAGCGGCCGG-TAGG, generated by Horizon Discovery. All mouse procedures were performed in accordance with University of Texas M.D. Anderson Cancer Center guidelines. Genomic DNA was iso-lated from tail biopsies and analyzed by PCR. The PRMT5 C terminus WT and ⌬HA allele were identified using the following oligonucleotides: 5Ј-CCGCCTGTGTCTTTCGTATT-3Ј and 5Ј-GTTGGCCACCATGACATTAG-3Ј. PCRs generated a 336 bp band in the wild-type allele and a 348 bp band in the PRMT5 ⌬HA allele. The PCR products for founder mice were sequenced to verify the fidelity of the mutation. ES cell lines were established as described previously (52).
Author Contributions-A. B. E. performed and analyzed the experiments shown in Figs. 1, 3, and 4. K. B. performed the experiments shown in Fig. 2. N. V., J. K., and T. C. generated ES cell lines. M. S. and C. W. provided reagents and were involved in the revision of the manuscript. G. G. and S. G. addressed the concerns of the reviewers. M. T. B. designed the research, directed the experiments, and wrote the paper. All authors reviewed the results and approved the final version of the manuscript.