Structural basis of O-GlcNAc recognition bymammalian 14-3-3 proteinsClifford A. Tolemana,1, Maria A. Schumachera,1, Seok-Ho Yub, Wenjie Zenga, Nathan J. Coxa, Timothy J. Smitha,Erik J. Soderblomc, Amberlyn M. Wandsb, Jennifer J. Kohlerb, and Michael Boycea,2

aDepartment of Biochemistry, Duke University School of Medicine, Durham, NC 27710; bDepartment of Biochemistry, University of Texas SouthwesternMedical Center, Dallas, TX 75390; and cDuke Proteomics and Metabolomics Core Facility, Center for Genomic and Computational Biology, Duke University,Durham, NC 27710

Edited by Carolyn R. Bertozzi, Stanford University, Stanford, CA, and approved April 23, 2018 (received for review December 24, 2017)

O-GlcNAc is an intracellular posttranslational modification that gov-erns myriad cell biological processes and is dysregulated in humandiseases. Despite this broad pathophysiological significance, thebiochemical effects of most O-GlcNAcylation events remain unchar-acterized. One prevalent hypothesis is that O-GlcNAc moieties maybe recognized by “reader” proteins to effect downstream signaling.However, no general O-GlcNAc readers have been identified, leav-ing a considerable gap in the field. To elucidate O-GlcNAc signalingmechanisms, we devised a biochemical screen for candidate O-GlcNAcreader proteins. We identified several human proteins, including 14-3-3 isoforms, that bind O-GlcNAc directly and selectively. We demon-strate that 14-3-3 proteins bind O-GlcNAc moieties in human cells, andwe present the structures of 14-3-3β/α and γ bound to glycopeptides,providing biophysical insights into O-GlcNAc-mediated protein–proteininteractions. Because 14-3-3 proteins also bind to phospho-serine andphospho-threonine, they may integrate information from O-GlcNAcand O-phosphate signaling pathways to regulate numerousphysiological functions.

O-GlcNAc | reader proteins | 14-3-3 | enolase | EBP1

O-linked β-N-acetylglucosamine (O-GlcNAc) is an abundantposttranslational modification (PTM) of serines and threo-

nines on nuclear, cytoplasmic, and mitochondrial proteins, gov-erning diverse biological processes (1–3). In mammals, O-GlcNAcis added by O-GlcNAc transferase (OGT) and removed by O-GlcNAcase (OGA), and is essential, as genetic ablation of ei-ther enzyme is lethal in mice (4–6). Aberrant O-GlcNAc cycling isalso implicated in numerous human diseases, including cancer (1,7–9), diabetes (10–12), and neurodegeneration (13–16).Despite this broad pathophysiological significance, the molec-

ular mechanisms of O-GlcNAc signaling are poorly understood.One prevailing hypothesis is that O-GlcNAc moieties, similar toother intracellular PTMs, may be recognized by “reader” proteinsthat specifically bind O-GlcNAc to effect downstream functions(17, 18). However, although some examples of O-GlcNAc-mediatedprotein–protein interactions are known (19–26), no generalO-GlcNAc readers have been identified, and no structure of anyO-GlcNAc-mediated protein–protein interaction has been reported,leaving the biochemical and biophysical basis of O-GlcNAc recog-nition unclear. To address this knowledge gap, we devised a screento discover candidate mammalian O-GlcNAc reader proteins. Wediscovered that multiple human proteins, including 14-3-3 iso-forms (17, 27, 28), bind selectively to O-GlcNAcylated substrates,and we determined structures of human 14-3-3β/α and γ boundto glycopeptides, providing atomic resolution information on anO-GlcNAc-mediated protein–protein interaction. Remarkably,the 14-3-3 O-GlcNAc-binding pocket overlaps with its well-knownphosphorylated ligand binding site (17, 27, 28). Therefore, 14-3-3 proteins may serve as bifunctional readers, integrating in-formation in the previously documented, extensive crosstalkbetween O-GlcNAcylation and O-phosphorylation pathways (1,29–33).

ResultsWe developed a biochemical approach to test the hypothesis thatO-GlcNAc is specifically recognized by mammalian reader pro-teins. First, we derived a consensus O-GlcNAcylated peptidesequence by aligning 802 mapped Ser-O-GlcNAc sites (34–36)(Fig. 1A) (www.phosphosite.org). We noted that a Pro-Val-Sertripeptide observed previously in smaller datasets (37, 38) alsoemerged in our sequence, suggesting that this motif may beimportant for O-GlcNAc modification and/or recognition. Wereasoned that this consensus peptide, when glycosylated, mightserve as biochemical “bait” for diverse O-GlcNAc reader pro-teins. Human OGT efficiently glycosylated a bait peptide con-taining the consensus sequence and a polyethylene glycol-biotinanchor (SI Appendix, Fig. S1A) with a single O-GlcNAc on theexpected serine, as determined by mass spectrometry (MS; Fig. 1Band SI Appendix, Fig. S1B). The GlcNAc-binding lectin wheatgerm agglutinin was affinity-enriched via pulldown with the baitglycopeptide, compared with the unglycosylated control peptide,even in the presence of excess nonspecific protein (Fig. 1C). Theseresults indicated that our approach could selectively capture GlcNAc-binding proteins.Using the bait glycopeptide and MS proteomics, we affinity-

captured and identified dozens of endogenous nuclear and cy-toplasmic proteins that selectively bound the O-GlcNAcylated

Significance

O-GlcNAc is an abundant, reversible posttranslational modification(PTM) of nuclear and cytoplasmic proteins in animals and plants.O-GlcNAc regulates a wide range of biological processes, and ab-errant O-GlcNAcylation is implicated in numerous human diseases.However, key aspects of O-GlcNAc signaling remain poorly un-derstood. For example, it is not known whether “reader” proteinsexist to recognize and bind to O-GlcNAc, as is true for many otherPTMs. We used a biochemical method to identify candidate humanO-GlcNAc reader proteins, and then characterized them at thebiochemical and biophysical levels. Our results address a signif-icant gap in the cell signaling field by revealing the biochemicaland structural basis for the recognition of O-GlcNAc by con-served human proteins.

Author contributions: C.A.T., M.A.S., S.-H.Y., J.J.K., and M.B. designed research; C.A.T.,M.A.S., S.-H.Y., N.J.C., T.J.S., E.J.S., and A.M.W. performed research; C.A.T., W.Z., andA.M.W. contributed new reagents/analytic tools; C.A.T., M.A.S., S.-H.Y., N.J.C., E.J.S.,J.J.K., and M.B. analyzed data; and C.A.T. and M.B. wrote the paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.

Published under the PNAS license.

Data deposition: The atomic coordinates and structure factors have been deposited in theProtein Data Bank, www.wwpdb.org (PDB ID codes 6BYJ, 6BYK, 6BZD, and 6BYL).1C.A.T. and M.A.S. contributed equally to this work.2To whom correspondence should be addressed. Email: michael.boyce@duke.edu.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1722437115/-/DCSupplemental.

Published online May 21, 2018.

5956–5961 | PNAS | June 5, 2018 | vol. 115 | no. 23 www.pnas.org/cgi/doi/10.1073/pnas.1722437115

Dow

nloa

ded

by g

uest

on

Sep

tem

ber

12, 2

020

probe in extracts from three different human cell lines (Fig. 1 Dand E and SI Appendix, Fig. S1C). Then, we inspected our MSdatasets for the proteins most enriched by the O-GlcNAcylatedbait in at least two cell types (SI Appendix, Fig. S1C). One proteinmeeting these criteria was importin-β1, which mediates nuclearcargo trafficking (39) and was previously shown to interact di-rectly with O-GlcNAc moieties on nuclear pore proteins in hu-man cells (40). We confirmed by immunoblot (IB) that importin-β1 and its homolog importin-5 were specifically enriched by thebait glycopeptide (Fig. 1F). Interestingly, several other proteinsenriched in our experiments exist as dimers or higher-order olig-omers, possibly indicating that avidity effects from the denselyglycoprotein-decorated beads contributed to their successful O-GlcNAc-specific purification (SI Appendix, Fig. S1C). These re-sults demonstrated that our method can identify authentic humanO-GlcNAc-interacting proteins.We selected three other glycopeptide-enriched proteins for

further study: α-enolase, ErbB3-binding protein (EBP1), and 14-3-3 (SI Appendix, Fig. S1C). α-enolase is a glycolytic enzyme (41),EBP1 inhibits cell growth and proliferation induced by the re-ceptor tyrosine kinase ErbB3 (42), and 14-3-3 proteins bindphosphoserine/threonine moieties to govern numerous processes,including growth factor signaling, mitotic exit, and apoptosis (17,27, 28). We verified that the bait glycopeptide specifically enrichedboth endogenous (Fig. 2A) and recombinant-purified (Fig. 2B)forms of human α-enolase, EBP1, and 14-3-3β/α and γ (repre-sentative paralogs), demonstrating that they bind O-GlcNAc di-rectly, and not through bridging proteins. We confirmed thisconclusion using fluorescence anisotropy (FA), which revealedsaturable binding of all three proteins to the glycopeptide, but notto the unglycosylated peptide (Fig. 2 C and D and SI Appendix,Fig. S2). Extending these observations, we found that humanα-enolase, EBP1, and 14-3-3 also directly bound a consensusglycopeptide derived from 676 mapped threonine-O-GlcNAc sites(34–36) (SI Appendix, Fig. S3) (www.phosphosite.org), and thatthe murine orthologs of all three proteins were also specificallyaffinity-enriched by bait glycopeptides (SI Appendix, Fig. S4).Together, these results demonstrate that mammalian α-enolase,

EBP1, and 14-3-3 share evolutionarily conserved O-GlcNAc-binding properties.To determine whether α-enolase, EBP1, and 14-3-3 bind O-

GlcNAc moieties in vivo, we created expression constructs encod-ing mCherry fused to the Ser-O-GlcNAc bait sequence and a

A

O-GlcNAc

tubulin

importin-5

importin- 1

O-GlcNAc

O-GlcNAc

B

C

nuclear cytoplasmicE

O-GlcNAc

Fsilver stain

D

Fig. 1. A biochemical strategy to identify O-GlcNAc reader proteins. (A) Peptide logo derived from 802 mammalian serine-O-GlcNAc sites. (B) Human OGT O-GlcNAcylates the bait peptide once at the central serine. MALDI spectra of the bait peptide after mock (Left) or complete (Right) O-GlcNAcylation reaction.m/z of modified peptide corresponds to +1 GlcNAc and +1 Mg2+. See also SI Appendix, Fig. S1B. (C) Fluorescein-tagged wheat germ agglutinin was incubatedwith glycopeptide or mock-modified peptide (±O-GlcNAc), without (Left) or with (Right) 105-fold excess BSA as a nonspecific protein control. Capturedmaterial was washed, eluted, and analyzed by SDS/PAGE and fluorescence scanning. (D) Nuclear and cytoplasmic 293T proteins were captured by glyco-peptide pulldown and analyzed by SDS/PAGE and silver stain. (E) Summary of MS proteomics results from samples shown in D. Red circles, proteins identifiedin unmodified peptide pulldowns (control). Blue (nuclear) and green (cytoplasmic) circles, proteins identified in glycopeptide pulldowns. Similar results wereobtained with HT1080 and Jurkat cells. See also SI Appendix, Fig. S1C. (F) Glycopeptide pulldowns from 293T extracts were analyzed by IB. Tubulin (notenriched in any MS sample) is a negative control.

14-3-3

O-GlcNAc -

mP

[14-3-3

glycopeptide

controlProtein KD

-enolase 183 ± 54EBP1 172 ± 8

14-3-3 324 ± 12

Binding affinities for selected protein/glycopeptide interactions

PP1 (MBP)

14-3-3 /

A

C

D

B

Fig. 2. Human protein binds O-GlcNAc directly and selectively. (A) Glyco-peptide pulldowns from 293T extracts were analyzed by IB. Tubulin is anegative control. (B) Recombinant-purified proteins were analyzed by gly-copeptide pulldown and IB. Protein phosphatase-1 (PP1) (not enriched in anyMS sample) is a negative control. (C) Fluorescein-labeled bait peptide wasmock- (blue) or O-GlcNAc-modified (red), and binding to 14-3-3γ was ana-lyzed by FA. Representative traces from triplicate experiment are shown. mP,millipolarization. (D) Dissociation constants for glycopeptide interactionswith α-enolase, EBP1, and 14-3-3γ, determined by FA.

Toleman et al. PNAS | June 5, 2018 | vol. 115 | no. 23 | 5957

BIOCH

EMISTR

Y

Dow

nloa

ded

by g

uest

on

Sep

tem

ber

12, 2

020

myc epitope. Endogenous 14-3-3, α-enolase, and EBP1 coim-munoprecipitated (IP-ed) with transfected mCherry-bait, butnot with a serine→alanine “mutant” bait (Fig. 3A). Moreover, aspecific small molecule inhibitor of OGT (5SGlcNAc) (43)greatly reduced the interaction between candidate reader pro-teins and mCherry-bait, whereas an OGA inhibitor (Thiamet-G) (44) had little effect (Fig. 3B). These results indicate thatconstitutive O-GlcNAcylation of the bait peptide sequence isefficient, and that endogenous α-enolase, EBP1, and 14-3-3 canbind O-GlcNAc moieties in human cells.To determine whether candidate reader proteins interact

with native OGT substrates in living cells, we used a chemicalbiology strategy (40) to covalently capture endogenous O-GlcNAc-mediated protein–protein interactions. Briefly, in thisapproach, human cells are metabolically labeled with a dia-zirine-bearing GlcNAc analog (GlcNDAz), which they convertto a nucleotide-sugar species used by OGT to modify its nat-ural substrates (40). Short UV treatment of GlcNDAz-labeledcells covalently crosslinks proteins within ∼2–4 Å of the O-GlcNDAz moiety, capturing direct binding partners but notnonspecific proteins (25, 26, 40). We performed GlcNDAzcrosslinking on human cells, IP-ed lysates with an anti-O-GlcNAcantibody, separated IP-ed proteins by SDS/PAGE, and usedquantitative proteomics to identify proteins within gel slices ofdefined molecular weight (MW) ranges (SI Appendix, Fig. S5).

Then, we inspected the results for proteins detected in a UV-dependent manner above their predicted MWs, indicatingO-GlcNDAz-dependent crosslinking. As expected, we observedUV-specific crosslinking in the 100–200-kDa range of bothknown O-GlcNAcylated proteins, such as nucleoporins and OGTitself (45–47), and known O-GlcNAc-interacting proteins, such asimportins (Fig. 3C) (40). Notably, we detected 14-3-3β/α, γ, e, andζ/δ (MW < 30 kDa) in the 100–200-kDa gel region exclusively inthe +UV sample (Fig. 3C). Similarly, α-enolase (MW 47 kDa)was enriched more than fivefold in the region above 200 kDa(SI Appendix, Fig. S5B). IBs confirmed the GlcNDAz-specificcrosslinking of 14-3-3β/α and γ to discrete endogenous humanproteins (Fig. 3D). Together, these results indicate that candidatereader proteins bind directly to O-GlcNAc moieties in vivo.Elegant studies of OGA–glycopeptide cocomplexes have pro-

vided mechanistic insight into O-GlcNAc hydrolysis (48, 49).However, no structure has been reported for an O-GlcNAc-mediated protein–protein interaction, leaving the biophysicalbasis of O-GlcNAc recognition unknown. To address this ques-tion, we focused on 14-3-3 proteins because of their role in cellsignaling through PTMs (17, 27, 50). O-GlcNAc binding is aconserved property of human 14-3-3 isoforms, because glyco-peptide pulldowns specifically enriched all family members, ex-cept σ, from mammalian cell extracts (Figs. 2A and 4A and SIAppendix, Fig. S1C). Moreover, glycopeptide pulldowns (Fig. 4B)and FA (Fig. 4C and SI Appendix, Fig. S6) with recombinant-purified protein confirmed the direct and selective binding oftwo additional 14-3-3 isoforms to O-GlcNAc, with no saturablebinding to the unglycosylated peptide detected. The affinities of14-3-3 isoforms for the bait glycopeptide (Fig. 2D and SI Ap-pendix, Fig. S6), although moderate, are comparable to thosereported previously for some phosphopeptide ligands fromphysiological binding partners (51, 52). To further test the po-tential physiological relevance of the affinities we observed for14-3-3 and model glycopeptides, we synthesized previously vali-dated 14-3-3-binding phosphopeptides from the cystic fibrosistransmembrane conductance regulator and the transcriptionfactor Snail (51, 52) (SI Appendix, Fig. S1A) and examined themunder identical FA assay conditions. Consistent with prior re-ports (51, 52), we observed weak binding with both cystic fibrosistransmembrane conductance regulator and Snail ligands, withaffinities lower than those we observed for the model glyco-peptides (SI Appendix, Fig. S7), demonstrating that the 14-3-3/glycopeptide affinities we observe are on par with some physio-logical 14-3-3/phosphopeptide interactions. By analogy to priorstudies comparing phosphopeptides and full-length phospho-proteins (53), we expect that 14-3-3 isoforms likely bind signifi-cantly tighter to native, full-length glycosylated partner proteinsthan they do to our consensus glycopeptides, which were designedto capture a variety of reader proteins, and hence were not se-lected or optimized for 14-3-3 binding in particular (Fig. 1A and SIAppendix, Fig. S3A).Next, to elucidate the biophysical basis of O-GlcNAc binding

by 14-3-3 paralogs, we obtained X-ray crystal structures of serineand threonine bait glycopeptides bound to 14-3-3β/α or γ (Fig. 4D and E and SI Appendix, Figs. S8–10). Strikingly, the glyco-peptides occupy the same amphipathic groove of 14-3-3, wherephosphorylated ligands bind (54). In addition, to protein/peptidebackbone contacts, 14-3-3 makes 10 hydrogen bonds with theglycan (Fig. 4 D and E), revealing the biophysical basis for O-GlcNAc-selective binding (Figs. 1–3). Moreover, the glyco-peptide and protein conformations were nearly identical inthe 14-3-3β/α and γ cocomplexes (Fig. 4 D, E, and I and SIAppendix, Fig. S9), providing a structural explanation for theconserved O-GlcNAc binding property of 14-3-3 isoforms (Fig.4A). Remarkably, the bound O-GlcNAc moiety is essentially su-perimposable on the O-phosphate group of prior phosphopeptide-bound structures of 14-3-3, with similar residues contacting bothPTMs, despite their considerable steric and electronic differences(Fig. 4F).

A

CD

B

Fig. 3. Candidate reader proteins bind O-GlcNAcylated substrates in humancells. (A) 293T cells were transfected with mCherry, mCherry-Ser-bait-myc,or mCherry-Ala-bait-myc. Lysates were analyzed by anti-myc IP and IB.(B) 293T cells were transfected as in A, incubated 24 h, and treated with vehicle,50 μM 5SGlcNAc (OGT inhibitor), or 50 μM Thiamet-G (OGA inhibitor) for 6 h.Lysates were analyzed by anti-myc IP and IB. The EBP1 and pan-14-3-3 signalin each IP lane was quantified and expressed as a percentage of the signalfrom the corresponding input sample (noted directly beneath each IB).(C) AGX1(F383G)-expressing HeLa cells (40) were treated with GlcNDAzprecursor and ultraviolet (or not, negative control). Lysates were analyzed byanti-O-GlcNAc IP, SDS/PAGE, and silver stain. Gel slices of defined MW rangeswere analyzed by spectral index quantitation MS proteomics (74). Graphdepicts selected proteins identified in the 100–200-kDa gel region andenriched more than fivefold in the +UV sample versus the −UV (control)sample. Y-axis indicates protein abundance, as quantified by mean ioncurrent (MIC). X-axis indicates the ratio (fold-enrichment) of each proteinin the +UV/−UV samples. Proteins identified in the +UV sample but absentfrom the −UV control are denoted as∞ enrichment. Axis breaks indicate scalechanges only. See also SI Appendix, Fig. S5. (D) AGX1(F383G)-expressing293T cells were transfected with vector or HA-14-3-3β/α or γ, treated asindicated, and UV-irradiated. Lysates were analyzed by anti-HA IB. Squares,uncrosslinked 14-3-3. Arrows, 14-3-3 crosslinks (i.e., O-GlcNAc-mediatedinteractions).

5958 | www.pnas.org/cgi/doi/10.1073/pnas.1722437115 Toleman et al.

Dow

nloa

ded

by g

uest

on

Sep

tem

ber

12, 2

020

We next used our structures to design mutations that uncouplethe O-GlcNAc and O-phosphate binding properties of 14-3-3.Mutating R57, R132, or Y133 (14-3-3γ numbering) to glutamateis known to abolish phospho-ligand binding because of electro-static repulsion (55, 56). We reasoned that these mutants mightstill bind uncharged O-GlcNAc moieties. Indeed, the R57E,R132E, and Y133E 14-3-3γ mutants selectively bound glyco-peptides in pulldown and FA experiments (Fig. 4G and H and SIAppendix, Fig. S6) comparably to wild type (Fig. 2). Further-more, we obtained the structure of a 14-3-3γ R57E-glycopeptidecomplex, which confirmed that O-GlcNAc binding is preservedwithout significant conformational changes (Fig. 4I). In a re-ciprocal experiment, we created N178Y and V181W mutants,which we predicted would not bind O-GlcNAc, hypothesizingthat the bulkier tyrosine or tryptophan residue would steri-cally disrupt the key contacts made by N178 and V181 to the

glycopeptide (Fig. 4 D and E). Indeed, both N178Y and V181W14-3-3γ failed to bind O-GlcNAcylated ligands in a pulldownassay (Fig. 4G). Together, these results provide biochemical andstructural insights into selective, O-GlcNAc-mediated protein–protein interactions, and reveal the biophysical basis of glycanbinding by 14-3-3 proteins, a property distinct and separable fromO-phosphate binding.Finally, we leveraged our structural and mutagenesis data to

further investigate the relationship between 14-3-3 and O-GlcNAc signaling in human cells. Consistent with an earlier re-port (57), we found that expressed wild-type 14-3-3γ was broadlydistributed throughout the nucleus and cytoplasm, whereas theR57E and Y133E 14-3-3γ mutants, which bind O-GlcNAc (Fig. 4G–I and SI Appendix, Fig. S6) but not O-phosphate (55, 56), localizespecifically to the nucleus (SI Appendix, Fig. S11). Moreover, bothmutants were dispersed throughout the cell on treatment with either

A

D E

F G

H I

B C

Fig. 4. Structural basis of O-GlcNAc recognition by 14-3-3 proteins. (A) 293T extracts were analyzed by glycopeptide pulldown and IB. (B and C) Recombinant-purified 14-3-3 proteins were analyzed by glycopeptide pulldown and IB (B) or FA (C). (D and E) Structures of 14-3-3γ (cyan) (D) or 14-3-3β/α (rose) (E) bound toglycopeptide (yellow peptide, green O-GlcNAc). Lines indicate protein-glycopeptide hydrogen bonds (Left), and blue mesh indicates the initial Fo-Fc map,calculated before glycopeptide addition, contoured at 2.9 σ (14-3-3γ) or 2.5 σ (14-3-3β/α) (Right). (F) Overlay of 14-3-3γ-glycopeptide (from D) and 14-3-3γ-phosphopeptide from tyrosine hydroxylase (magenta protein, red phosphopeptide; PDB 4J6S) (75). O-GlcNAc and O-phosphate moieties are highlighted bytransparent surfaces. (G) Recombinant-purified wild-type or R57E, Y133E, N178Y, or V181W mutant 14-3-3γ was analyzed by glycopeptide pulldown and IB.(H) 14-3-3γ R57E binding to glycopeptide was analyzed by FA. (I) Overlays of structures of 14-3-3γ (cyan), 14-3-3β/α (green), and 14-3-3γ R57E (magenta),displaying nearly identical modes of glycopeptide binding.

Toleman et al. PNAS | June 5, 2018 | vol. 115 | no. 23 | 5959

BIOCH

EMISTR

Y

Dow

nloa

ded

by g

uest

on

Sep

tem

ber

12, 2

020

of two structurally unrelated small molecule OGT inhibitors (SIAppendix, Fig. S11), suggesting that their localization depends onendogenous glycosylated (but not phosphorylated) nuclear ligands.These data further support a role for 14-3-3 as O-GlcNAc readerproteins in live human cells.

DiscussionDespite their importance, the biochemical mechanisms of O-GlcNAc signaling remain incompletely understood. To addressthis knowledge gap, we tested the hypothesis that the O-GlcNAcmodification is specifically recognized by mammalian readerproteins. Several pioneering studies from the Lefebvre grouppreviously reported that Hsp70 family chaperones bind GlcNAc-agarose resin and associate with O-GlcNAcylated mammalianproteins in a nutrient- and stress-responsive manner (58–62).However, whether Hsp70 chaperones or other human proteinsserve as direct and specific O-GlcNAc-binding readers remainedunclear.Our results demonstrate that diverse human proteins bind O-

GlcNAcylated substrates directly and selectively. For example,α-enolase binds O-GlcNAcylated substrates (Figs. 1–3) and isitself O-GlcNAcylated (8), indicating that glycosylation maygovern glycolysis through regulated protein–protein interactions.In support of this possibility, our glycopeptide pulldowns enrichedseveral glycolytic and pentose phosphate pathway enzymes (SIAppendix, Fig. S1C), consistent with the well-documented roleof O-GlcNAc signaling in nutrient sensing (1, 2). Similarly,EBP1 binds O-GlcNAcylated ligands (Figs. 1–3) and interactswith known OGT substrates, including the transcriptional regulatorsE2F1 and Sin3A (63–65), suggesting that O-GlcNAcylation mayinfluence ErbB3 signaling through regulated multiprotein complexesof EBP1. The structural basis and functional consequences ofO-GlcNAc binding by α-enolase and EBP1 are currently underinvestigation.The 14-3-3 isoforms, also identified in our screen as O-

GlcNAc binders, play central roles in many physiological pro-cesses, including mitogenic signaling, cell cycle progression, andcell death (17, 27, 28). Our biophysical studies with 14-3-3 pro-vide structures of O-GlcNAc-mediated protein–protein in-teraction, lending insight into the signaling mechanisms of 14-3-3and O-GlcNAc alike. On the basis of these results, we suggestthat 14-3-3 proteins are well-suited to serve as general O-GlcNAc readers. Our structures reveal that the amphipathicgroove of 14-3-3 is especially well equipped to bind the Pro-Val-Ser/Thr motif (Fig. 1A and SI Appendix, Fig. S3) present inhundreds of natural OGT substrates (37, 38). Specifically, theproline in this glycopeptide motif occupies a shallow hydropho-bic pocket of 14-3-3 that likely would not tolerate large or polarside chains (Fig. 4 and SI Appendix, Fig. S9). Similarly, the valine,although surface exposed, lies near the conserved L225 and theCβ atoms of D228 and N229 (14-3-3γ numbering), making sev-eral favorable contacts (Fig. 4 and SI Appendix, Fig. S9). Im-portantly, specific 14-3-3 residues, (e.g., D129, N178, and E185,14-3-3γ numbering) participate in O-GlcNAc binding but not O-phosphate binding (Fig. 4 and SI Appendix, Fig. S9), consistentwith our observation that the O-GlcNAc and O-phosphatebinding properties of 14-3-3 isoforms are genetically separable(Fig. 4 G–I and SI Appendix, Fig. S9) (55, 56). Our results alsodemonstrate that multiple 14-3-3 isoforms from different mam-malian species bind Ser- and Thr-O-GlcNAc moieties specificallyand directly (Figs. 2–4 and SI Appendix, Figs. S2–S4), likelythrough nearly identical biophysical contacts (Fig. 4I and SIAppendix, Fig. S9). Therefore, the amphipathic groove of mam-malian 14-3-3 isoforms is an evolutionarily conserved O-GlcNAc-binding protein module, in addition to its well-establishedO-phosphate-binding role.O-GlcNAc and O-phosphate often compete for identical or

nearby residues on numerous substrates, producing a complexfunctional interplay between these PTMs (1, 29–33). Our resultsraise the possibility that 14-3-3 isoforms act as bifunctionalreader proteins of specific O-GlcNAcylated or phosphorylated

ligands. Moreover, because 14-3-3 isoforms homo- and hetero-dimerize (17, 27, 28), the two amphipathic grooves of a single14-3-3 dimer could conceivably bind one O-GlcNAc and one O-phosphate, respectively, when interacting with the many proteinsknown to be both phosphorylated and glycosylated (1, 29–33).Indeed, some native binding partners of 14-3-3 are phosphory-lated at multiple sites, providing both high- and lower-affinityphospholigands for binding 14-3-3 dimers, permitting combina-torial tuning of cell signaling through multiple PTMs on a singleprotein (52, 66, 67). We speculate that O-GlcNAc moieties maysimilarly serve as high- or lower-affinity 14-3-3 binding sites,perhaps in combination with phosphorylation sites, on endoge-nous human proteins. Testing these hypotheses will be an im-portant goal for future studies.The natural O-GlcNAcylated binding partners of 14-3-3 pro-

teins remain to be identified. However, our proteomics resultsrevealed several known endogenous nuclear glycoproteins in thesame GlcNDAz-crosslinked, high-molecular-weight complexesas endogenous 14-3-3 (Fig. 3C and SI Appendix, Fig. S5). Im-portantly, the chemical mechanism and short crosslinking radiusof the GlcNDAz reagent ensure that these in vivo interactionsare mediated by direct binding to glycans (40). These data sug-gest that 14-3-3 may exploit high O-GlcNAc ligand avidity tobind directly to heavily glycosylated nucleoporins or other nuclearOGT substrates (68–70). Consistent with this hypothesis, ourmicroscopy experiment with wild-type and nonphosphobindingmutants of 14-3-3γ also suggest that 14-3-3 proteins bind en-dogenous nuclear O-GlcNAc moieties in human cells (SI Ap-pendix, Fig. S11).Characterization of the major O-GlcNAcylated binding part-

ners of 14-3-3 will likely require extensive proteomic and cellbiological efforts. As a first step toward this goal, we cloned wild-type and R57E mutant 14-3-3γ with tandem HA and Halo tagsand performed GlcNDAz crosslinking in 293T cells. Consistentwith our prior results (Fig. 3 C and D), wild-type HA-Halo-14-3-3γ crosslinked in a GlcNDAz-specific fashion (SI Appendix, Fig.S12). Importantly, R57E mutant HA-Halo-14-3-3γ exhibitedGlcNDAz crosslinking nearly identical to that of wild-type pro-tein (SI Appendix, Fig. S12), further reinforcing the conclusionthat 14-3-3γ binds endogenous OGT substrates in a phosphory-lation-independent manner. In future work, we anticipate thatHA-Halo-14-3-3 constructs will allow stringent affinity purifica-tion and proteomic analyses of GlcNDAz-crosslinked proteins,revealing the major endogenous O-GlcNAcylated ligands of14-3-3.In summary, we report the systematic identification of candi-

date reader proteins that bind O-GlcNAc directly and specifically.In addition, our structures of 14-3-3/glycopeptide complexes pro-vide a biophysical characterization of an O-GlcNAc-mediatedinteraction, laying the groundwork for future functional studies.The biochemical and structural characterization of reader proteinsof other PTMs has greatly advanced our understanding of cellsignaling (17, 18, 71–73). We expect that our results will similarlyfacilitate new analyses of 14-3-3 proteins and other candidateO-GlcNAc readers in a wide range of biological contexts.

Materials and MethodsFull proteomics datasets are available in Datasets S1–S5. X-ray crystal struc-tures have been deposited in the Protein Data Bank under ID codes 6BYJ,6BYK, 6BZD and 6BYL.

Additional details are available in the SI Appendix, Supporting Materialsand Methods.

ACKNOWLEDGMENTS. We thank Suzanne Walker (Harvard Medical School)and Pei Zhou (Duke University School of Medicine) for chemicals and plasmids,Benjamin Swarts (Central Michigan University) for Ac45SGlcNAc, the Ad-vanced Light Source beamline 8.3.1 for data collection, and James Alvarez,Jen-Tsan Ashley Chi, Benjamin Swarts, and members of the M.B. laboratoryfor helpful suggestions. This work was supported by a Rita Allen Founda-tion Scholar Award and NIH Grants 1R01GM118847-01 and UL1TR001117(to M.B.), and grants from the NIH (R21DK112733) and Welch Foundation(I-1686) (to J.J.K.).

5960 | www.pnas.org/cgi/doi/10.1073/pnas.1722437115 Toleman et al.

Dow

nloa

ded

by g

uest

on

Sep

tem

ber

12, 2

020

1. Hart GW, Slawson C, Ramirez-Correa G, Lagerlof O (2011) Cross talk between O-GlcNAcylation and phosphorylation: Roles in signaling, transcription, and chronicdisease. Annu Rev Biochem 80:825–858.

2. Bond MR, Hanover JA (2013) O-GlcNAc cycling: A link between metabolism andchronic disease. Annu Rev Nutr 33:205–229.

3. Yang X, Qian K (2017) Protein O-GlcNAcylation: Emerging mechanisms and functions.Nat Rev Mol Cell Biol 18:452–465.

4. Shafi R, et al. (2000) The O-GlcNAc transferase gene resides on the X chromosome andis essential for embryonic stem cell viability and mouse ontogeny. Proc Natl Acad SciUSA 97:5735–5739.

5. Keembiyehetty C, et al. (2015) Conditional knockout reveals a requirement for O-GlcNAcase in metabolic homeostasis. J Biol Chem 290:7097–7113.

6. Yang YR, et al. (2012) O-GlcNAcase is essential for embryonic development andmaintenance of genomic stability. Aging Cell 11:439–448.

7. Ma Z, Vosseller K (2013) O-GlcNAc in cancer biology. Amino Acids 45:719–733.8. Yi W, et al. (2012) Phosphofructokinase 1 glycosylation regulates cell growth and

metabolism. Science 337:975–980.9. Singh JP, Zhang K, Wu J, Yang X (2015) O-GlcNAc signaling in cancer metabolism and

epigenetics. Cancer Lett 356:244–250.10. Vaidyanathan K, Wells L (2014) Multiple tissue-specific roles for the O-GlcNAc post-

translational modification in the induction of and complications arising from type IIdiabetes. J Biol Chem 289:34466–34471.

11. Hardivillé S, Hart GW (2014) Nutrient regulation of signaling, transcription, and cellphysiology by O-GlcNAcylation. Cell Metab 20:208–213.

12. Ma J, Hart GW (2013) Protein O-GlcNAcylation in diabetes and diabetic complications.Expert Rev Proteomics 10:365–380.

13. Yuzwa SA, et al. (2012) Increasing O-GlcNAc slows neurodegeneration and stabilizestau against aggregation. Nat Chem Biol 8:393–399.

14. Vaidyanathan K, Durning S, Wells L (2014) Functional O-GlcNAc modifications: Implicationsin molecular regulation and pathophysiology. Crit Rev Biochem Mol Biol 49:140–163.

15. Yuzwa SA, Vocadlo DJ (2014) O-GlcNAc and neurodegeneration: Biochemical mechanismsand potential roles in Alzheimer’s disease and beyond. Chem Soc Rev 43:6839–6858.

16. Zhu Y, Shan X, Yuzwa SA, Vocadlo DJ (2014) The emerging link between O-GlcNAcand Alzheimer disease. J Biol Chem 289:34472–34481.

17. Reinhardt HC, Yaffe MB (2013) Phospho-Ser/Thr-binding domains: Navigating the cellcycle and DNA damage response. Nat Rev Mol Cell Biol 14:563–580.

18. Wozniak GG, Strahl BD (2014) Hitting the ‘mark’: Interpreting lysine methylation inthe context of active transcription. Biochim Biophys Acta 1839:1353–1361.

19. Gewinner C, et al. (2004) The coactivator of transcription CREB-binding protein in-teracts preferentially with the glycosylated form of Stat5. J Biol Chem 279:3563–3572.

20. Fujiki R, et al. (2011) GlcNAcylation of histone H2B facilitates its monoubiquitination.Nature 480:557–560.

21. Ruan HB, et al. (2012) O-GlcNAc transferase/host cell factor C1 complex regulatesgluconeogenesis by modulating PGC-1α stability. Cell Metab 16:226–237.

22. Xiao J, et al. (2014) O-GlcNAc-mediated interaction between VER2 and TaGRP2 elicitsTaVRN1 mRNA accumulation during vernalization in winter wheat. Nat Commun 5:4572.

23. Chen PH, et al. (2017) Glycosylation of KEAP1 links nutrient sensing to redox stresssignaling. EMBO J 36:2233–2250.

24. Tarbet HJ, Toleman CA, Boyce M (2018) A sweet embrace: Control of protein-proteininteractions by O-linked β-N-acetylglucosamine. Biochemistry 57:13–21.

25. Cox NJ, et al. (2018) Dynamic glycosylation governs the vertebrate COPII proteintrafficking pathway. Biochemistry 57:91–107.

26. Tarbet HJ, et al. (2018) Site-specific glycosylation regulates the form and function ofthe intermediate filament cytoskeleton. eLife 7:e31807.

27. Freeman AK, Morrison DK (2011) 14-3-3 proteins: Diverse functions in cell pro-liferation and cancer progression. Semin Cell Dev Biol 22:681–687.

28. Neal CL, Yu D (2010) 14-3-3ζ as a prognostic marker and therapeutic target for cancer.Expert Opin Ther Targets 14:1343–1354.

29. Wang Z, et al. (2010) Extensive crosstalk between O-GlcNAcylation and phosphory-lation regulates cytokinesis. Sci Signal 3:ra2.

30. Wang S, et al. (2012) Extensive crosstalk between O-GlcNAcylation and phosphory-lation regulates Akt signaling. PLoS One 7:e37427.

31. Trinidad JC, et al. (2012) Global identification and characterization of both O-GlcNA-cylation and phosphorylation at the murine synapse. Mol Cell Proteomics 11:215–229.

32. Zhong J, et al. (2015) Quantitative phosphoproteomics reveals crosstalk between phos-phorylation and O-GlcNAc in the DNA damage response pathway. Proteomics 15:591–607.

33. Leney AC, El Atmioui D, Wu W, Ovaa H, Heck AJR (2017) Elucidating crosstalkmechanisms between phosphorylation and O-GlcNAcylation. Proc Natl Acad Sci USA114:E7255–E7261.

34. Hornbeck PV, Chabra I, Kornhauser JM, Skrzypek E, Zhang B (2004) PhosphoSite: Abioinformatics resource dedicated to physiological protein phosphorylation.Proteomics 4:1551–1561.

35. Hornbeck PV, et al. (2012) PhosphoSitePlus: A comprehensive resource for in-vestigating the structure and function of experimentally determined post-trans-lational modifications in man and mouse. Nucleic Acids Res 40:D261–D270.

36. Hornbeck PV, et al. (2015) PhosphoSitePlus, 2014: Mutations, PTMs and recalibrations.Nucleic Acids Res 43:D512–D520.

37. Vosseller K, et al. (2006) O-linked N-acetylglucosamine proteomics of postsynapticdensity preparations using lectin weak affinity chromatography and mass spec-trometry. Mol Cell Proteomics 5:923–934.

38. Chalkley RJ, Thalhammer A, Schoepfer R, Burlingame AL (2009) Identification ofprotein O-GlcNAcylation sites using electron transfer dissociation mass spectrometryon native peptides. Proc Natl Acad Sci USA 106:8894–8899.

39. Cook A, Bono F, Jinek M, Conti E (2007) Structural biology of nucleocytoplasmictransport. Annu Rev Biochem 76:647–671.

40. Yu SH, et al. (2012) Metabolic labeling enables selective photocrosslinking of O-GlcNAc-modified proteins to their binding partners. Proc Natl Acad Sci USA 109:4834–4839.

41. Díaz-Ramos A, Roig-Borrellas A, García-Melero A, López-Alemany R (2012) α-Enolase,a multifunctional protein: Its role on pathophysiological situations. J BiomedBiotechnol 2012:156795.

42. Hamburger AW (2008) The role of ErbB3 and its binding partners in breast cancerprogression and resistance to hormone and tyrosine kinase directed therapies.J Mammary Gland Biol Neoplasia 13:225–233.

43. Gloster TM, et al. (2011) Hijacking a biosynthetic pathway yields a glycosyltransferaseinhibitor within cells. Nat Chem Biol 7:174–181.

44. Yuzwa SA, et al. (2008) A potent mechanism-inspired O-GlcNAcase inhibitor thatblocks phosphorylation of tau in vivo. Nat Chem Biol 4:483–490.

45. Hanover JA, Cohen CK, Willingham MC, Park MK (1987) O-linked N-acetylglucos-amine is attached to proteins of the nuclear pore. Evidence for cytoplasmic and nu-cleoplasmic glycoproteins. J Biol Chem 262:9887–9894.

46. Holt GD, et al. (1987) Nuclear pore complex glycoproteins contain cytoplasmicallydisposed O-linked N-acetylglucosamine. J Cell Biol 104:1157–1164.

47. Kreppel LK, Blomberg MA, Hart GW (1997) Dynamic glycosylation of nuclear andcytosolic proteins. Cloning and characterization of a unique O-GlcNAc transferasewith multiple tetratricopeptide repeats. J Biol Chem 272:9308–9315.

48. Schimpl M, Borodkin VS, Gray LJ, van Aalten DM (2012) Synergy of peptide and sugarin O-GlcNAcase substrate recognition. Chem Biol 19:173–178.

49. Li B, Li H, Lu L, Jiang J (2017) Structures of human O-GlcNAcase and its complexesreveal a new substrate recognition mode. Nat Struct Mol Biol 24:362–369.

50. Milroy LG, Brunsveld L, Ottmann C (2013) Stabilization and inhibition of protein-protein interactions: The 14-3-3 case study. ACS Chem Biol 8:27–35.

51. Stevers LM, et al. (2016) Characterization and small-molecule stabilization of themultisite tandem binding between 14-3-3 and the R domain of CFTR. Proc Natl AcadSci USA 113:E1152–E1161.

52. Hou Z, et al. (2010) 14-3-3 binding sites in the snail protein are essential for snail-mediated transcriptional repression and epithelial-mesenchymal differentiation.Cancer Res 70:4385–4393.

53. Sluchanko NN, et al. (2017) Structural basis for the interaction of a human small heatshock protein with the 14-3-3 universal signaling regulator. Structure 25:305–316.

54. Yaffe MB, et al. (1997) The structural basis for 14-3-3:phosphopeptide binding spec-ificity. Cell 91:961–971.

55. Petosa C, et al. (1998) 14-3-3zeta binds a phosphorylated Raf peptide and an un-phosphorylated peptide via its conserved amphipathic groove. J Biol Chem 273:16305–16310.

56. Wang H, Zhang L, Liddington R, Fu H (1998) Mutations in the hydrophobic surface ofan amphipathic groove of 14-3-3zeta disrupt its interaction with Raf-1 kinase. J BiolChem 273:16297–16304.

57. Brunet A, et al. (2002) 14-3-3 transits to the nucleus and participates in dynamicnucleocytoplasmic transport. J Cell Biol 156:817–828.

58. Lefebvre T, et al. (2001) Identification of N-acetyl-d-glucosamine-specific lectins from ratliver cytosolic and nuclear compartments as heat-shock proteins. Biochem J 360:179–188.

59. Guinez C, Lemoine J, Michalski JC, Lefebvre T (2004) 70-kDa-heat shock proteinpresents an adjustable lectinic activity towards O-linked N-acetylglucosamine.Biochem Biophys Res Commun 319:21–26.

60. Guinez C, Losfeld ME, Cacan R, Michalski JC, Lefebvre T (2006) Modulation of HSP70GlcNAc-directed lectin activity by glucose availability and utilization.Glycobiology 16:22–28.

61. Guinez C, et al. (2007) Hsp70-GlcNAc-binding activity is released by stress, proteasomeinhibition, and protein misfolding. Biochem Biophys Res Commun 361:414–420.

62. Guinez C, et al. (2010) Arginine 469 is a pivotal residue for the Hsc70-GlcNAc-bindingproperty. Biochem Biophys Res Commun 400:537–542.

63. Yang X, Zhang F, Kudlow JE (2002) Recruitment of O-GlcNAc transferase to promotersby corepressor mSin3A: Coupling protein O-GlcNAcylation to transcriptional re-pression. Cell 110:69–80.

64. Wells L, Slawson C, Hart GW (2011) The E2F-1 associated retinoblastoma-susceptibilitygene product is modified by O-GlcNAc. Amino Acids 40:877–883.

65. Zhang Y, Akinmade D, Hamburger AW (2005) The ErbB3 binding proteinEbp1 interacts with Sin3A to repress E2F1 and AR-mediated transcription. NucleicAcids Res 33:6024–6033.

66. Kostelecky B, Saurin AT, Purkiss A, Parker PJ, McDonald NQ (2009) Recognition of anintra-chain tandem 14-3-3 binding site within PKCepsilon. EMBO Rep 10:983–989.

67. Mils V, et al. (2000) Specific interaction between 14-3-3 isoforms and the humanCDC25B phosphatase. Oncogene 19:1257–1265.

68. Li B, Kohler JJ (2014) Glycosylation of the nuclear pore. Traffic 15:347–361.69. Ruba A, Yang W (2016) O-GlcNAc-ylation in the nuclear pore complex. Cell Mol

Bioeng 9:227–233.70. Eustice M, Bond MR, Hanover JA (2017) O-GlcNAc cycling and the regulation of nu-

cleocytoplasmic dynamics. Biochem Soc Trans 45:427–436.71. Sawicka A, Seiser C (2014) Sensing core histone phosphorylation–A matter of perfect

timing. Biochim Biophys Acta 1839:711–718.72. Wilkinson AW, Gozani O (2014) Histone-binding domains: Strategies for discovery

and characterization. Biochim Biophys Acta 1839:669–675.73. Chi P, Allis CD, Wang GG (2010) Covalent histone modifications–Miswritten, mis-

interpreted and mis-erased in human cancers. Nat Rev Cancer 10:457–469.74. Trudgian DC, et al. (2011) Comparative evaluation of label-free SINQ normalized

spectral index quantitation in the central proteomics facilities pipeline. Proteomics 11:2790–2797.

75. Skjevik AA, et al. (2014) The N-terminal sequence of tyrosine hydroxylase is a con-formationally versatile motif that binds 14-3-3 proteins and membranes. J Mol Biol426:150–168.

Toleman et al. PNAS | June 5, 2018 | vol. 115 | no. 23 | 5961

BIOCH

EMISTR

Y

Dow

nloa

ded

by g

uest

on

Sep

tem

ber

12, 2

020