A systematic study of modulation of ADAM-mediatedectodomain shedding by site-specific O-glycosylationChristoffer K. Gotha, Adnan Halima, Sumeet A. Khetarpalb,c, Daniel J. Raderb,c, Henrik Clausena,and Katrine T.-B. G. Schjoldagera,1

aCopenhagen Center for Glycomics, Department of Cellular and Molecular Medicine, Faculty of Health Sciences, University of Copenhagen, Blegdamsvej3, DK-2200 Copenhagen N, Denmark; bDepartment of Genetics, Perelman School of Medicine, University of Pennsylvania, Philadelphia, PA 19104; andcDepartment of Medicine, Perelman School of Medicine, University of Pennsylvania, Philadelphia, PA 19104

Edited by Monty Krieger, Massachusetts Institute of Technology, Cambridge, MA, and approved October 9, 2015 (received for review June 8, 2015)

Regulated shedding of the ectodomain of cell membrane proteinsby proteases is a common process that releases the extracellulardomain from the cell and activates cell signaling. Ectodomainshedding occurs in the immediate extracellular juxtamembraneregion, which is also where O-glycosylation is often found andexamples of crosstalk between shedding and O-glycosylation havebeen reported. Here, we systematically investigated the potentialof site-specific O-glycosylation mediated by distinct polypeptideGalNAc-transferase (GalNAc-T) isoforms to coregulate ectodomainshedding mediated by the A Disintegrin And Metalloproteinase(ADAM) subfamily of proteases and in particular ADAM17. Weanalyzed 25 membrane proteins that are known to undergoADAM17 shedding and where the processing sites included Ser/Thr residues within ± 4 residues that could represent O-glycosites.We used in vitro GalNAc-T enzyme and ADAM cleavage assays todemonstrate that shedding of at least 12 of these proteins arepotentially coregulated by O-glycosylation. Using TNF-α as anexample, we confirmed that shedding mediated by ADAM17 iscoregulated by O-glycosylation controlled by the GalNAc-T2 iso-form both ex vivo in isogenic cell models and in vivo in mouseGalnt2 knockouts. The study provides compelling evidence for awider role of site-specific O-glycosylation in ectodomain shedding.

ACE | IL-6RA | ErbB4 | ADAM10 | TACE

Ectodomain shedding, i.e., the proteolytic release of the ex-tracellular domain of membrane proteins is a very common

and highly regulated process, in which specific proteases selectivelycleave substrate proteins in juxtamembrane regions. Ectodomainshedding is usually regulated in response to appropriate stimuli andaffects a large number of different classes of proteins including cellmembrane receptors, growth factors, cytokines, and cell adhesionmolecules (1). The shedding event may induce different effects in-cluding activation, inactivation, or modulation of the functionalproperties of proteins (2) and are important in regulating a numberof biological processes such as signal transduction, adhesion, pro-liferation and differentiation (1, 3). Ectodomain shedding is pre-dominantly mediated by Zinc-dependent metalloproteinases, andone of the key classes of proteases involved is the ADAM subfamily.Members of the ADAM family are type 1 transmembrane proteinsthat share conserved domains, including a prodomain that serves asan inhibitor of the proteolytic activity; a metalloprotease domainthat may or may not contain an active site with the conserved motif(HEXGHXXGXXHD) common for Zinc-binding proteases; adisintegrin domain and a cysteine-rich region, which are bothbelieved to be involved in substrate recognition; an EGF-likedomain; a transmembrane domain; and a cytoplasmic domainimportant for intracellular signaling. Dysregulation of ectodo-main shedding is involved in a variety of diseases including in-flammation, cardiovascular diseases, diabetes, Alzheimer disease,and cancer (1, 4). However, the underlying mechanisms for dys-regulation are poorly understood. Defined consensus motifs forADAM cleavage sites have not been identified, but for a number ofmembrane proteins undergoing regulated ectodomain shedding the

cleavage sites and the specific ADAM proteases involved have beenidentified (5). Interestingly, cleavage sites in the extracellular juxta-membrane region are often found in close proximity to O-glycosyl-ation sites, hence glycosylation may affect susceptibility for proteolyticprocessing of proteins (6).GalNAc-type O-glycosylation (hereafter simply O-glycosylation)

is found on the majority of proteins passing through the secretorypathway and often in linker and stem regions of proteins (7). Thistype of protein glycosylation is unique in being differentiallyregulated by a large family of up to 20 distinct polypeptide GalNAc-transferases (GalNAc-Ts), which initiate O-glycosylation bycatalyzing addition of the first GalNAc residue to Ser and Thrresidues (and possibly Tyr residues) (8). These GalNAc-T isoformshave different yet partly overlapping peptide acceptor substratespecificities, and have different expression patterns in cells andtissues, which opens for a high degree of differential and possiblydynamic regulation of the O-glycoproteome and individualO-glycosites in proteins (8). Previously it has been shown thatsite-specific O-glycosylation specifically inhibits Furin proproteinprocessing ± 3 residues of cleavage sites (9), and in generalO-glycosylation is known to confer structure and stability to stemregions of proteins and preventing proteolysis. Classic studiesdemonstrated the importance of O-glycosylation in the stem regionof the low-density-lipoprotein (LDL) receptor using the CHO ldlDcell system that enables analysis of the general effect of O-glyco-sylation (10). A number of other examples have confirmed this,and extended it to related receptors (11–13). Recently, it wasshown that mutational loss of O-glycosylation in the stem regionof the Interleukin-2 receptor subunit alpha (IL2Rα) result inincreased shedding of the ectodomain (14). However, it is currently

Significance

Ectodomain shedding is a central event in a range of biologicalprocesses and pathways, however the underlying mechanismsare still not fully understood. Here we present evidence thatsite-specific O-glycosylation regulated by individual GalNAc-Transferase isoforms, serve to coregulate ectodomain shed-ding, predominantly through blocking of cleavage. Our generalfinding is exemplified by the specific role of a single GalNAc-Tisoform (GalNAc-T2) in coregulating TNF-alpha release in vitro,ex vivo in isogenic cell models, and in vivo in mouse Galnt2knockouts. Adding the large family of GalNAc-T isoforms toregulation of ectodomain shedding substantially increase theability to fine-tune this important process on a substrate level.

Author contributions: C.K.G., S.A.K., D.J.R., H.C., and K.T.-B.G.S. designed research; C.K.G.,A.H., and S.A.K. performed research; C.K.G., A.H., S.A.K., and K.T.-B.G.S. analyzed data;and C.K.G., H.C., and K.T.-B.G.S. wrote the paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.1To whom correspondence should be addressed. Email: Schjoldager@sund.ku.dk.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1511175112/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1511175112 PNAS | November 24, 2015 | vol. 112 | no. 47 | 14623–14628

CELL

BIOLO

GY

unclear to what extent O-glycosylation in the stem region serves aprotective role for proteolytic processing in general or more spe-cific and regulated roles in ectodomain shedding by, e.g., ADAMs.O-glycosylation has been proposed to specifically affect

ADAM-mediated ectodomain shedding (15, 16). Tumor ne-crosis factor alpha (TNF-α), a type 2 transmembrane precursorprotein that undergoes ectodomain shedding to produce a solubletruncated form, was shown to be O-glycosylated near a knownADAM cleavage site in lymphoblastic leukemia B cells (16), andMinond and colleagues showed that glycosylation of synthetic pep-tides covering the TNF-α cleavage site affected in vitro ADAMprocessing (15). Furthermore, it was recently found that also ADAMmediated cleavage of the extracellular N-terminal tail of the GPCRβ1-adrenergic receptor may be affected by O-glycosylation (17). Wehypothesized that important roles of site-specific O-glycosylationwould have to be carried out by individual GalNAc-T isoforms toenable tuneable coregulation. This hypothesis leaves the substratespecificities of GalNAc-T isoforms and their unique functions as acentral element in deciphering the potential for O-glycosylation tocoregulate ectodomain shedding. This function was recently ex-emplified with the GalNAc-T11 isoform identified as a candidatedisease gene in new-born heart heterotaxy (18). Thus, in vitrostudies suggested that the GalNAc-T11 enzyme exhibited selec-tive substrate specificity for the juxtamembrane region of Notch1and coregulated ADAM17 processing of Notch1 (19).In the present study, we have systematically investigated the

potential of site-specific O-glycosylation mediated by distinctGalNAc-T isoforms to coregulate known ADAM processing. Wechose ADAM17 (also known as TACE) as a representative ex-ample identified 25 membrane proteins with known ADAM17processing sites in juxtamembrane regions and putative proximalO-glycosylation sites defined as Ser/Thr residues within ± 4residues of the processing site. Synthetic peptides covering suchregions from approximately half of the analyzed proteins wereO-glycosylated by one or more GalNAc-T isoforms and in mostcases glycosylation modulated in vitro ADAM cleavage. We usedTNF-α as an example to demonstrate that shedding by ADAM17is coregulated by site-specific O-glycosylation controlled by theGalNAc-T2 isoform ex vivo and in vivo. Our study provides evi-dence for a wider role of site-specific O-glycosylation in coregulationof ectodomain shedding.

ResultsScreening for Site-Specific O-Glycosylation Adjacent to Known ADAMCleavage Sites.ADAM17 is the best-characterized member of theADAM family with respect to membrane protein ectodomaincleavage sites, so we chose to focus on ADAM17 to investigatethe potential role of site-specific O-glycosylation. We data minedthe literature for experimentally validated ADAM17 cleavagesites with Ser/Thr residues ± 4 amino acid residues of thecleavage sites, and identified 25 cleavage sites from 23 differentmembrane proteins (Fig. 1 and Fig. S1). We then designed short(20mer) peptides covering the cleavage and putative adjacentglycosylation sites, and used in vitro enzyme time-course analysesmonitored by MALDI-TOF mass spectrometry to evaluate poten-tial for O-glycosylation with a panel of recombinant GalNAc-Ts.We tested 10 human recombinant GalNAc-Ts covering the majorisoforms expressed in most cells and found that 12 of the 25designed peptide substrates served as substrates for O-glycosylationby one or more of the enzymes (Fig. 1 and Fig. S1). Consider-able data indicate that in vitro GalNAc-T enzyme analysis ofshort peptide substrates correlates well with in vivo glycosylationin cells (7, 20), and in a large scale analysis of peptides designed tocover known O-glycosites about 40% were not glycosylated (20).This finding suggests, if anything, that in vitro analysis appearsto underestimate O-glycosylation capacity (20), and hence thatadditional candidates among the 25 identified in fact may beglycosylated in vivo.

Among the 12 candidates for combined cleavage and glyco-sylation were six membrane receptors: TGF-beta receptor type-1(TGFR-1), Receptor tyrosine-protein kinase erbB-4 (ErbB4),Interleukin-6 receptor subunit alpha (IL6-RA), Tumor necrosisfactor receptor superfamily member 1B (TNF-R2), IFN alpha/betareceptor 2 (IFN-R-2) and Growth hormone receptor (GHR); twoproteases: ADAM10 and Angiotensin-converting enzyme (ACE);two growth factors: Betacellulin (BTC) and Proheparin-bindingEGF-like growth factor (HB-EGF); as well as the cytokine TNF-α,and Myelin basic protein (MBP). Interestingly, the peptide sub-strates designed for TNF-α, TGFR-1, ErbB4, and HB-EGFwere only glycosylated at one or two sites by a single GalNAc-Tisoform (Fig. 1). Several peptide substrates were glycosylated bytwo GalNAc-T isoforms (IFN-R2, p75, and ACE), whereas thepeptides for BTC, MBP, ADAM10, IL6-RA, GHR, and ACEwere glycosylated at one or more sites by multiple GalNAc-Ts(Fig. 1). Thus, the first group of proteins has potential for beingdifferentially O-glycosylated by a single GalNAc-T isoform, butthe second group may also be regulated in cells where subsets ofGalNAc-T isoforms are expressed.

Site-Specific O-Glycosylation Modulates in Vitro ADAM Processing.To evaluate the effect of GalNAc O-glycosylation on ADAMprocessing of the 12 candidate peptides we analyzed in vitroADAM cleavage of peptides and glycopeptides produced by therelevant GalNAc-Ts. The O-glycosites of these glycopeptideswere characterized by ESI-LIT-FT-MS, and five recombinanthuman ADAM proteases (ADAM8, -9, -10, -12, and -17) weretested in in vitro enzyme assays, where cleavage product for-mation was monitored by time-course MALDI-TOF MS as de-scribed (21). This assay provides a relative estimate of theefficiencies of ADAMs in cleaving the (GalNAc)-peptides in vitro,and more detailed analysis of kinetic properties of the ADAMswith these substrates is beyond this study. Ten out of 12 peptideswere cleaved in vitro by one or more ADAM isoforms and, amongthe corresponding glycopeptides, 8 of 10 cleavages by one or moreADAMs were affected by glycosylation (Fig. 2). In most cases,glycosylation inhibited cleavage (TNF-α, ADAM10, GHR, IL-6R,ErbB4, HB-EGF, and ACE), and in one case glycosylation en-hanced cleavage (ErbB4). For TNF-α glycosylation of S80 resultedin partial protection from ADAM17, -9, and -10 processing at76A↓V77, whereas ADAM12 processing at 79S↓S80 was completelyblocked (Figs. 2 and 3). For ADAM10 T625, glycosylation blockedcleavage at 621G↓R622 (Fig. 2). Glycosylation of T250 and S255 in

Fig. 1. In vitro O-glycosylation analysis of known ADAM substrates.(A) Peptides of 20–28 residues covering published ADAM cleavagesites. O-glycosylated residues are indicated by GalNAc (square). (B) In vitroO-glycosylation with GalNAc-T1, -T2, -T3, -T4, -T5, -T11, -T12, -T13, -T14,and -T16. The number of incorporated sites is shown in parenthesis.

14624 | www.pnas.org/cgi/doi/10.1073/pnas.1511175112 Goth et al.

GHR blocked ADAM9 and -17 cleavages at 252P↓Q253. Glycosyl-ation of T352, S359, and S360 in IL6-RA blocked ADAM9 cleavageat 359P↓V360. IFN-R2 glycosylation at T228 blocked ADAM9 cleav-age at 228T↓L229. Cleavage of ACE by ADAM17 at 1232R↓S1233 wasblocked by glycosylation at S1230 and S12333, and glycosylation ofHB-EGF at T75 and T85 blocked ADAM17 processing at 72L↓R72

and 72R↓V73; no difference in efficiency of the two cleavage siteswas observed. In ErbB4, glycosylation at T643 and T650 led to in-creased cleavage by ADAM10 at 637P↓W638, and the ADAM17cleavage sites shifted from 641H↓S642 and from 645P↓Q646 to 637P↓W638,no difference in cleavage efficiency of 645P↓Q646 and 637P↓W638

was observed. In addition, the ADAM9 cleavage at 641H↓S642 wasblocked and ADAM8 cleavage at 645P↓Q646 was unaffected (Fig. 2).We did not observe any effect of glycosylation on cleavage of p75(ADAM17 and -9 at 261P↓V262) orMBP (ADAM17 at 202H↓Y203, andADAM8, -10, and -12 at 207P↓Q208) (Fig. 2).

Ectodomain Shedding of TNF-α Is Coregulated by GalNAc-T2. Weselected TNF-α as a model for ex vivo validation in cells becauseO-glycosylation has been speculated to coregulate the sheddingof TNF-α and because it served as an example of GalNAc-Tisoform-specific (GalNAc-T2) glycosylation (Fig. 3 and Fig. S1B).TNF-α is produced as a type II transmembrane precursor (26 kDa)that undergoes regulated ectodomain shedding of a 17-kDaC-terminal form that assembles into functional homotrimers andenters the bloodstream (22). To probe the role of glycosylation forTNF-α shedding ex vivo we expressed an alkaline phosphatase-tagged juxtamembrane TNF-α reporter construct (Fig. 4A) in apanel of isogenic cell lines with and without specific GalNAc-Ts.Human liver HepG2 isogenic cells without GalNAc-T1 (HepG2ΔT1)or GalNAc-T2 (HepG2ΔT2), as well as the latter with GalNAc-T2reintroduced (HepG2ΔT2+T2) were transiently transfected for 48 husing the TNF-α reporter construct, and total cell lysates were

analyzed by Western blotting (Fig. 4B). Comparing HepG2ΔT2 toWT and ΔT1 isogenic cells, we observed specific increased shed-ding in T2 KO cells, which were abrogated when HepG2ΔT2clones were rescued by site-directed AAVS1 insertion of GalNAc-T2 (Fig. 4B). The increased shedding was verified in four differentHepG2ΔT2 clones and in two GalNAc-T2 rescue clones in threeindependent experiments (Fig. 4B and Fig. S2).To further confirm this finding and exclude cell-specific factors,

we also tested the TNF-α reporter construct in a panel of HEK293isogenic cell lines (HEK293 WT, HEK293ΔT2, and HEK293ΔT3),which express a different repertoire of GalNAc-Ts with the majordifference being GalNAc-T3. Similar to HepG2 cells we found se-lective increased shedding of TNF-α (∼40%) in the HEK293ΔT2clones, confirming that GalNAc-T2 may play a critical role forprocessing and shedding of the TNF-α reporter construct inagreement with the in vitro analysis (Fig. 4 C and D).

LPS-Induced Plasma Levels of TNF-α Is Elevated in T2 KO MiceCompared with WT. We next tested the in vivo effect of com-plete GalNAc-T2 deficiency on levels of circulating TNF-α afterlipopolysaccharide (LPS) stimulation. We compared TNF-α re-lease in WT and Galnt2−/− KO mice. Animals were injectedusing 0.5 mg/kg body weight of LPS by IP injection and plasmalevels were measured by ELISA at 0, 1-h, 3-h, and 6-h timepoints. Galnt2−/− mice showed a statistically significant increasein TNF-α plasma levels at all measured time points (1 h, 3 h, and6 h) (Fig. 4E), confirming that GalNAc-T2 plays a role in TNF-αrelease and regulation. Furthermore, we observed no differencein the plasma levels of IL6-RA before or after LPS induction inthese animals (Fig. S2). Finally, expression analysis showed thatTNF-α expression levels were not changed between WT and KOanimals (Fig. S2).

Congenital Gene Variants with Potential Dysregulated EctodomainShedding due to Loss of O-Glycosylation. Several diseases or syn-dromes have been linked to deficiencies in GalNAc-T isoformsby various genetic association studies. The molecular mecha-nisms behind these associations are in most cases unresolved, but

Fig. 2. Summary of in vitro ADAM cleavage of naked peptides and glyco-peptides. Peptide sequences covering known ADAM processing sites (arrowsand numbers indicate cleavage sites and ADAM isoforms, respectively).ADAM8, -9, -10, -12, and -17 were tested on all peptides that were in vitroO-glycosylated. Numbers in red indicate that we observed protection fromcleavage after glycosylation, numbers in black indicate no effect, and num-bers in green indicate that cleavage was increased. No cleavage was observedof BTC and TGFR1. *The ADAM17 cleavage of ERBb4 shifted from H642↓STLP↓Qof the naked peptide to 637P↓W of the glycopeptide, and ERBb4 glycosylationincreased cleavage by ADAM10.

A

C

B

D

Fig. 3. TNF-α is selectively O-glycosylated by GalNAc-T2, and this modulatesin vitro ADAM cleavage. (A) Schematic drawing of TNF-ectodomain shed-ding zooming in on the juxtamembrane part that undergoes cleavage withindicated cleavage and glycosylation sites. (B) In vitro glycosylation of thejuxtamembrane part of TNF-α with GalNAc-T2. Horizontal arrow indicatesmass shift due to GalNAc glycosylation. (C) In vitro cleavage analysis of TNF-αpeptide and glycopeptide by ADAM17 monitored by MALDI-TOF. Productsformed after 30 min, 1 h, and 4 h are shown. ADAM17 cleaved at 76A↓V.Cleavage of the glycopeptide was less efficient. (D) ADAM12 cleaved at 79S↓S.Cleavage of the glycopeptide was almost completely abolished. ADAM9 and-10 had similar cleavage patterns as ADAM17 (Fig. 2). ADAM8 was not ob-served to cleave the peptide. Sequence, cleavage site (scissors), and site ofglycosylation (square) are shown above each spectra.

Goth et al. PNAS | November 24, 2015 | vol. 112 | no. 47 | 14625

CELL

BIOLO

GY

deficiencies in distinct GalNAc-T isoform functions have beenshown to cause diseases due to lack of coregulation of proteolyticprocessing by site-specific O-glycosylation (19, 23, 24). Inline with these observations, we considered whether any ofthe O-glycosites potentially involved in coregulation of ectodomainshedding identified in this study were found to be mutated indisease. We therefore searched for known missense SNVs as-sociated with phenotypes or functional change of the variantprotein that were located in the juxtamembrane regions in closeproximity to our newly identified glycosylation sites presentedhere (±5 amino acids).We identified two missense SNVs in 2 (ACE and IL6-RA) of

the 10 proteins with putative O-glycosites affecting ADAMprocessing within a distance of ±5 residues of the processing site.Interestingly, these two variants of ACE and IL6-RA (P1228Land D358A, respectively) have been found to be associated withincreased serum levels of their shed ectodomains (25, 26). Totest whether these variations in amino acid sequence affectedin vitro O-glycosylation and/or ADAM processing, we testedpeptides with the substitutions. As shown in Fig. S3B, the gly-cosylation efficiency by GalNAc-T3 of the ACE P1228L variantpeptide was dramatically decreased. However, the IL6-RAD358A variant was a better substrate for GalNAc-T1, GalNAc-T2, and GalNAc-T3, i.e., more peptide were glycosylated of thevariant compared with the WT when comparing the glycosylationreactions at different time points (1 h, 2 h, 4 h) (Fig. S3B). Wealso observed an additional fourth site added by GalNAc-T3,which was not detected before (Fig. S3B). The amino acid substi-tution did not seem to affect the substrate efficiency for GalNAc-T11 and GalNAc-T13. To test if the amino acid substitutions aloneaffected ADAM activity, we analyzed in vitro cleavage of the non-glycosylated variant peptides. We found that the ACE variant pep-tide was cleaved with comparable efficiency as the WT peptidewithout glycosylation (Fig. S3C), indicating that it was not theamino acid substitution itself, but rather the loss of O-glycosylation,that is responsible for increased shed ACE levels in individuals withthis gene variant. Conversely, the naked IL6-RA variant peptide

was much more efficiently cleaved, which is likely to account forthe increased serum levels in people carrying this genotype.Nevertheless, the glycosylation was still found to inhibit ADAMmediated cleavage of the IL6-RA peptide. Our results highlightthe importance of considering the glycosylation status of the jux-tamembrane part of a protein when investigating shedding and ac-tivation of membrane receptors.Finally, we tested the artificially produced ErbB4 mutant Q646C,

which has been reported to induce a constitutively active receptorpredicted to undergo constitutive dimerization (27). ErbB4 is alsoknown to undergo Regulated Intramembrane Proteolysis (RIP)leading to intracellular signaling (28), and this mechanism couldalso play a role in the increased receptor activity. We tested themutant Q646C peptide by in vitro assays and observed a dramaticreduction in the glycosylation compared with the WT peptide(Fig. S3B).

DiscussionO-glycosylation in the juxtamembrane region of cell membraneproteins have for long been known to confer general stability andextended structure to the stem region that may be important forprotrusion of the receptor domains (6). However, the extent towhich isolated distinct sites of O-glycosylation play specific rolesfor regulated proteolytic processing events such as ectodomainshedding has remained unclear. Moreover, provided distinctO-glycosites serve such functions it seems conceivable that theglycosylation of these sites would be directed by distinct mem-bers of the large GalNAc-T family to enable reasonable regu-lation. Here, through systematic analysis of membrane proteinswith known ADAM17 processing sites, we demonstrate thatproximal site-specific O-glycosylation is likely to serve a commoncoregulatory role in ectodomain shedding. We further demon-strate that most of the identified O-glycosites were controlled byonly one or a few GalNAc-T isoforms, which opens the possi-bility for selective regulation through tuning of the expression ofspecific GalNAc-T isoforms. We selectively studied ADAM17processing sites, but it is likely that these findings can be ex-trapolated to other proteases involved in ectodomain shedding,stressing the need to include analysis of potential O-glycosylationadjacent to processing sites in future studies of ectodomainshedding. We also demonstrate the importance of consideringO-glycosylation when analyzing the effects SNVs may exert.GalNAc-type O-glycosylation is highly abundant with perhaps

over 85% of all proteins trafficking the secretory pathway beingO-glycosylated, and interestingly a majority of the identifiedO-glycoproteins only have very few O-glycosites in isolated positions(7). Although our knowledge of the O-glycoproteome has greatlyexpanded in recent years with the introduction of differentO-glycoproteomics strategies (29, 30), there are still technical dif-ficulties in identifying, for example, O-glycosites close to the jux-tamembrane region due to lack of efficient proteolytic cleavagesites for bottom-up mass spectrometry sequencing strategies (31).Although prediction algorithms for O-glycosylation are available(7), these may not have sufficient examples from juxtamembraneregions to serve as good predictors for O-glycosites in this region.Another obstacle is our limited knowledge on the substratespecificities of individual GalNAc-T isoforms and in particular thespecific O-glycosites they control nonredundantly in cells. Onlywith such information will it be possible to define the regulatablepart of the O-glycoproteome. Progress has been made with dif-ferential O-glycoproteomes using isogenic SimpleCells with andwithout individual GalNAc-T isoforms (32) and systematic in vitroanalysis of O-glycosylation (20), but we are still far from being ableto identify O-glycosites that are being regulated by individualGalNAc-Ts. We therefore used in vitro GalNAc-T enzyme assaysto identify putative O-glycosylation sites, which revealed potentialO-glycosites within ±4 residues of almost 50% of the analyzedADAM17 cleavage sites and glycosylation of most of these

A B

DC E

Fig. 4. Ectodomain shedding of TNF-α is coregulated by GalNAc-T2 in Cellline models and in vivo. (A) Depiction of the TNF-α reporter constructcontaining GalNAc-T2 specific glycosylation and ADAM cleavage sites.(B) Western blot visualizing cleavage patterns in HepG2 WT, ΔT2, ΔT1, andΔT2+T2 cells transiently transfected (48 h) with TNF-α reporter construct.(C) Visualization of AP activity in SDS-polyacrylamide gel in cell lysates andsupernatants fromHEK293wild type andΔT2 of full-length reporter and cleavedfragment. (D) Relative levels of shedding in HEK293 wild type, ΔT2, and ΔT3.Alkaline activity was measured in media and lysate. (E) Plasma levels (pg/mL) ofTNF-α after LPS induction in WT and GalNAc-T2 KO mice. Statistical analyseswere performed using the Student’s t test. Error bars show SDs.

14626 | www.pnas.org/cgi/doi/10.1073/pnas.1511175112 Goth et al.

affected in vitro ADAM cleavage (Fig. 2). Interestingly, in mostcases glycosylation affected cleavage negatively, but in one casecleavage was positively affected, similar to what we have pre-viously reported (19). It is important to note that we only testedthe simplest O-glycoform with just a GalNAc residue attached,whereas most cells will produce larger structures mainly of core1structure (Galβ1–3GalNAcα1-O-Ser/Thr) with sialic acid cap-ping. We have, however, previously shown that the mere pres-ence of an O-glycan is sufficient to evaluate an effect onproprotein processing (21), but also that the complete sialylatedcore1 structure may exert more pronounced effects (23).In agreement with our starting hypothesis, we also found that the

majority of the glycosites identified that modulated ADAM17 pro-cessing in vitro were glycosylated in a GalNAc-T isoform-specificmanner. The majority of the O-glycoproteome is covered by func-tional redundancy among GalNAc-T isoforms, and the widelyexpressed isoforms, GalNAc-T1, T2, and T3, appear to cover themajority of the O-glycosylation events in cells, whereas otherGalNAc-T isoforms have more restricted and specialized functions(8, 20). It is therefore striking that most of the O-glycosites pre-dicted to modulate ADAM17 ectodomain shedding events werecontrolled by nonredundant GalNAc-T isoform-specific functions,further supporting that they likely play coregulatory roles in ecto-domain shedding. In four cases (IFNAR2, ADAM10, GHR, andBTC) the O-glycosites were potentially controlled by two or threeGalNAc-T isoforms with overlapping functions (Fig. 1), however,because most cells express only a limited subset of GalNAc-T iso-forms these glycosites may still be regulated by individual isoforms.In this respect, it is important to include the repertoire of GalNAc-Tsexpressed in a given cell when considering redundancies ofGalNAc-T isoform activities (8, 32). HB-EGF was specifically gly-cosylated by GalNAc-T3 at two residues, which blocked ADAM17cleavage (Figs. 1 and 2). Given that GalNAc-T3 is not expressed innormal liver (24), the hepatic form of HB-EGF is likely not gly-cosylated in these sites, but GalNAc-T3 is widely expressed in othercells and can be strongly up-regulated in disease (33). HB-EGF hasbeen shown to play a protective role in acute liver damage (34).Similarly GHR was glycosylated at one site by GalNAc-T1 and attwo sites by GalNAc-T3 and GalNAc-T13, suggesting that theremay be some degree of redundancy in control of O-glycosylation ofthese sites. However, again, GalNAc-T3 is not expressed in liver andGalNAc-T13 is predominantly expressed in the brain and a fewother tissues (8), so it is plausible that GalNAc-T1 alone controlsglycosylation of GHR in liver. These examples underline the com-plexity in deciphering the roles site-specific O-glycosylation mayexert on biological processes such as ectodomain shedding.Precise gene editing tools offer unprecedented options to

dissect and validate nonredundant biological functions of gly-cosyltransferase isoenzymes (35). The knockout strategy used inthis study overcome knockdown strategies issue with insufficientreduction in enzyme levels to produce substantial changes inglycosylation. We used a panel of ZFN edited isogenic cell linesto demonstrate that ectodomain shedding of TNF-α is specificallycoregulated by GalNAc-T2 site-specific glycosylation at Ser80

and a ZFN edited mouse model Galnt2−/− to validate thisin vivo. TNF-α is a central mediator in several diseases includingrheumatoid arthritis and Crohn’s disease (36), and blockage ofTNF-α is a well established treatment regime for these and otherconditions (37). Several regulatory mechanisms of TNF-α releaseand ectodomain shedding have been proposed (38–40), but theregulation and the importance of the different mechanisms stillremain poorly understood. The coregulatory role of GalNAc-T2offers new insight to the molecular mechanisms behind TNF-αrelease and may be important to consider in diseases with elevatedsecreted TNF-α levels. GALNT2 play additional important rolesin dyslipidaemia (24), and future studies need to address the ex-tent to which these functions are regulated and fine-tuned.

A number of diseases are associated with altered levels of shedcell membrane receptor ectodomains in serum, and in somecases mutations in the juxtamembrane region of these receptorshave been identified. More recently, somatic mutations in thejuxtamembrane region of the colony-stimulating factor 3 re-ceptor (CSF3R) that appear to eliminate O-glycosylation hasbeen identified as driver mutations in chronic neutrophilic leu-kemia (41). In this case, the mechanism was proposed to involveenhanced homodimerization of the receptor due to loss ofO-glycosylation and constitutive signaling. The data presentedhere provide an additional scenario where mutations may alter ef-ficiency of site-specific O-glycosylation, which subsequently affectthe proteolytic cleavage event and ectodomain shedding. Oneilluminating example may be ACE in which a SNV (P1228L) isassociated with 2.5-fold higher levels of shed circulating ACE(25). Elevated plasma ACE activity have been suggested to beassociated with increased risk of cardiovascular disease (42),although a later report has not been able to reproduce this effect(43). Similarly, an IL6-RA SNV (D358A) has also been linked tohigher soluble levels of the receptor (44). We found that the D358Amutation enhanced O-glycosylation and also enhanced ADAM17cleavage (Fig. S3C). Interestingly, site-specific O-glycosylationappears to play a role in coregulating both variants and thechange in IL6-RA signaling caused by the D358A SNVhas been suggested to have a pathogenic role in asthma and isassociated with more severe disease (45). Finally, an experi-mental mutation in ErbB4 in close proximity to the glycosylationsites identified here has previously been shown to result in amarked increase in signaling (27). Again, this mutation affectedefficiency of site-specific O-glycosylation (Fig. S3B). It is possiblethat the loss of O-glycosylation affects the signaling throughmodulating the ectodomain shedding and/or leads to increase indimerization and ligand independent signaling as recently de-scribed for the CSF3R (41).In summary, we present evidence that site-specific O-glycosyl-

ation regulated by distinct GalNAc-T isoforms may widely serveto coregulate ADAM mediated ectodomain shedding. Although,the role of glycosylation predominantly appears to reduce shed-ding, examples of enhancement of shedding are also found. Site-specific O-glycosylation has potential for complex modulation ofectodomain shedding to regulate efficiency and direction of pro-cessing site and protease selectivity. Adding the many GalNAc-Tisoforms to regulation of ectodomain shedding substantially in-crease the ability to fine-tune the selectivity of this importantprocess. Moreover, as demonstrated, disease-associated muta-tions in the juxtamembrane region of membrane proteins mayindirectly affect ectodomain shedding by affecting glycosylationefficiency. Our study clearly underlines the importance of con-sidering O-glycosylation and the GalNAc-T repertoire as co-regulators of ectodomain shedding

Experimental ProceduresGlycosyltransferase Assays. Recombinant glycosyltransferases were expressedas soluble secreted truncated proteins in insect cells (45). In vitro activityassays for GalNAc-T glycosylation of peptides (Schafer-N, NeoBioSci) wereperformed as described and monitored with MALDI-TOF (20).

Characterization of O-Glycosylation Sites by Electron Transfer Dissociation-MS2.Products of O-glycosylation reactions were characterized by electrospray ion-ization-linear ion trap-Fourier transform mass spectrometry (ESI-LIT-FT-MS) inan LTQ-Orbitrap XL hybrid spectrometer (Thermo Scientific) equipped forelectron transfer dissociation for peptide sequence analysis by MS/MS (MS2)with retention of glycan site-specific fragments (SI Experimental Procedures).

ADAM Cleavage of Peptides and Glycopeptides. In vitro ADAM cleavage ac-tivity were assayed by adding 150–600 nM ADAM17, -10 (Enzo Life Sciences),-8, -9 (provided by Alexander Matthias Kotzsch, NNF Center for ProteinResearch, Faculty of Health, University of Copenhagen, Copenhagen), or -12(a gift from the Kveiborg group, BRIC, Faculty of Health University of

Goth et al. PNAS | November 24, 2015 | vol. 112 | no. 47 | 14627

CELL

BIOLO

GY

Copenhagen), using 10 μg of peptide or glycopeptide substrate in a totalvolume of 25 μL (SI Experimental Procedures).

Precise Gene Edited Cell Models. To confirm the in vitro assay findings, wedeveloped cell line models capable of probing both the effect on cleavage bysite-specific glycosylation and the GalNAc-T isoform involved. Because GalNAc-Tsare differentially expressed we used two different cell lines, HepG2 and HEK293,representing different GalNAc-T repertoires. We used Zinc finger nuclease (ZFN)gene editing to knockout and/or knock in GALNT1 and GALNT2 (Sigma-Aldrich)in HepG2 as described (32) (SI Experimental Procedures).

TNF-α Shedding Assay. HepG2 WT, ΔT1, ΔT2, and T2 rescue and HEK293 WT,ΔT1, and ΔT2 were plated in six-well plates and transiently transfected withthe juxtamembrane-TNF-α-AP expression construct (generously provided byCarl Blobel, Cornell University, Ithaca, NY). The TNF-α reporter constructconsists of a N-terminal intracellular FLAG tag fused to the first 87 aminoacids of TNF-α, followed by alkaline phosphatase and a MYC tag. The full-length fusion protein has a size of ∼70.2 kDa, and the shed part is ∼60.4 kDa.

The construct has been described in details by Zheng and colleagues (46) (SIExperimental Procedures).

T2 KO Mouse and Measurement of TNF-α Plasma Levels After LPS Induction. T2KO male mice (n = 4) and WT male controls (n = 8) were administered LPSreconstituted in PBS at a dose of 0.5 mg/kg body weight by IP injection intotal volume of 250 μL. Blood was collected via retroorbital bleed before LPSadministration and at 1, 3, and 6 h after LPS injection. Plasma TNF-α wasmeasured from all timepoints by a mouse TNF-α ELISA (Quantikine ELISA,R&D Systems). All procedures were approved by the Institutional AnimalCare and Use Committee of the University of Pennsylvania.

ACKNOWLEDGMENTS. This work was supported by the Danish ResearchCouncils (including Sapere Aude Research Talent Grant to K.T.-B.G.S.),a program of excellence from the University of Copenhagen, The NovoNordisk Foundation, a program of excellence from the University ofCopenhagen (CDO2016), and The Danish National Research Foundation(DNRF107).

1. Weber S, Saftig P (2012) Ectodomain shedding and ADAMs in development.Development 139(20):3693–3709.

2. Schlöndorff J, Blobel CP (1999) Metalloprotease-disintegrins: Modular proteins ca-pable of promoting cell-cell interactions and triggering signals by protein-ectodo-main shedding. J Cell Sci 112(Pt 21):3603–3617.

3. Blobel CP (2005) ADAMs: Key components in EGFR signalling and development. NatRev Mol Cell Biol 6(1):32–43.

4. Arribas J, Esselens C (2009) ADAM17 as a therapeutic target in multiple diseases. CurrPharm Des 15(20):2319–2335.

5. Lange PF, Overall CM (2011) TopFIND, a knowledgebase linking protein termini withfunction. Nat Methods 8(9):703–704.

6. Varki A (1993) Biological roles of oligosaccharides: All of the theories are correct.Glycobiology 3(2):97–130.

7. Steentoft C, et al. (2013) Precision mapping of the human O-GalNAc glycoproteomethrough SimpleCell technology. EMBO J 32(10):1478–1488.

8. Bennett EP, et al. (2012) Control of mucin-type O-glycosylation: A classification of thepolypeptide GalNAc-transferase gene family. Glycobiology 22(6):736–756.

9. Schjoldager KT, Clausen H (2012) Site-specific protein O-glycosylation modulatesproprotein processing - deciphering specific functions of the large polypeptide Gal-NAc-transferase gene family. Biochim Biophys Acta 1820(12):2079–2094.

10. Kingsley DM, Kozarsky KF, Segal M, Krieger M (1986) Three types of low density li-poprotein receptor-deficient mutant have pleiotropic defects in the synthesis ofN-linked, O-linked, and lipid-linked carbohydrate chains. J Cell Biol 102(5):1576–1585.

11. Kozarsky K, Kingsley D, Krieger M (1988) Use of a mutant cell line to study the kineticsand function of O-linked glycosylation of low density lipoprotein receptors. Proc NatlAcad Sci USA 85(12):4335–4339.

12. May P, Bock HH, Nimpf J, Herz J (2003) Differential glycosylation regulates processingof lipoprotein receptors by gamma-secretase. J Biol Chem 278(39):37386–37392.

13. Magrané J, Casaroli-Marano RP, Reina M, Gåfvels M, Vilaró S (1999) The role ofO-linked sugars in determining the very low density lipoprotein receptor stability orrelease from the cell. FEBS Lett 451(1):56–62.

14. Karabasheva D, Cole NB, Donaldson JG (2014) Roles for trafficking and O-linkedglycosylation in the turnover of model cell surface proteins. J Biol Chem 289(28):19477–19490.

15. Minond D, et al. (2012) Discovery of novel inhibitors of a disintegrin and metal-loprotease 17 (ADAM17) using glycosylated and non-glycosylated substrates. J BiolChem 287(43):36473–36487.

16. Takakura-Yamamoto R, Yamamoto S, Fukuda S, Kurimoto M (1996) O-glycosylatedspecies of natural human tumor-necrosis factor-alpha. Eur J Biochem 235(1-2):431–437.

17. Hakalahti AE, et al. (2010) Human beta1-adrenergic receptor is subject to constitutiveand regulated N-terminal cleavage. J Biol Chem 285(37):28850–28861.

18. Fakhro KA, et al. (2011) Rare copy number variations in congenital heart diseasepatients identify unique genes in left-right patterning. Proc Natl Acad Sci USA 108(7):2915–2920.

19. Boskovski MT, et al. (2013) The heterotaxy gene GALNT11 glycosylates Notch to or-chestrate cilia type and laterality. Nature 504(7480):456–459.

20. Kong Y, et al. (2015) Probing polypeptide GalNAc-transferase isoform substratespecificities by in vitro analysis. Glycobiology 25(1):55–65.

21. Schjoldager KT, et al. (2011) A systematic study of site-specific GalNAc-type O-glycosyl-ation modulating proprotein convertase processing. J Biol Chem 286(46):40122–40132.

22. McGeehan GM, et al. (1994) Regulation of tumour necrosis factor-alpha processing bya metalloproteinase inhibitor. Nature 370(6490):558–561.

23. Kato K, et al. (2006) Polypeptide GalNAc-transferase T3 and familial tumoral calci-nosis. Secretion of fibroblast growth factor 23 requires O-glycosylation. J Biol Chem281(27):18370–18377.

24. Schjoldager KT, et al. (2010) O-glycosylation modulates proprotein convertase acti-vation of angiopoietin-like protein 3: Possible role of polypeptide GalNAc-trans-ferase-2 in regulation of concentrations of plasma lipids. J Biol Chem 285(47):36293–36303.

25. Kramers C, et al. (2001) Point mutation in the stalk of angiotensin-converting enzymecauses a dramatic increase in serum angiotensin-converting enzyme but no cardio-vascular disease. Circulation 104(11):1236–1240.

26. Reich D, et al.; Health, Aging and Body Composition (Health ABC) Study (2007) Ad-mixture mapping of an allele affecting interleukin 6 soluble receptor and interleukin6 levels. Am J Hum Genet 80(4):716–726.

27. Penington DJ, Bryant I, Riese DJ, 2nd (2002) Constitutively active ErbB4 and ErbB2mutants exhibit distinct biological activities. Cell Growth Differ 13(6):247–256.

28. Rio C, Buxbaum JD, Peschon JJ, Corfas G (2000) Tumor necrosis factor-alpha-convertingenzyme is required for cleavage of erbB4/HER4. J Biol Chem 275(14):10379–10387.

29. Halim A, Nilsson J, Ruetschi U, Hesse C, Larson G (2012) Human urinary glyco-proteomics; attachment site specific analysis of N- and O-linked glycosylations by CIDand ECD. Mol Cell Proteomics 11(4):M111 013649.

30. Steentoft C, et al. (2011) Mining the O-glycoproteome using zinc-finger nuclease-glycoengineered SimpleCell lines. Nat Methods 8(11):977–982.

31. Levery SB, et al. (2015) Advances in mass spectrometry driven O-glycoproteomics.Biochim Biophys Acta 1850(1):33–42.

32. Schjoldager KT, et al. (2012) Probing isoform-specific functions of polypeptide Gal-NAc-transferases using zinc finger nuclease glycoengineered SimpleCells. Proc NatlAcad Sci USA 109(25):9893–9898.

33. Taniuchi K, et al. (2011) Overexpression of GalNAc-transferase GalNAc-T3 promotespancreatic cancer cell growth. Oncogene 30(49):4843–4854.

34. Takemura T, et al. (2013) Conditional knockout of heparin-binding epidermal growthfactor-like growth factor in the liver accelerates carbon tetrachloride-induced liverinjury in mice. Hepatology Res 43(4):384–393.

35. Steentoft C, et al. (2014) Precision genome editing: A small revolution for glycobiology.Glycobiology 24(8):663–680.

36. Vassalli P (1992) The pathophysiology of tumor necrosis factors. Annu Rev Immunol10:411–452.

37. Tracey D, Klareskog L, Sasso EH, Salfeld JG, Tak PP (2008) Tumor necrosis factor an-tagonist mechanisms of action: A comprehensive review. Pharmacol Ther 117(2):244–279.

38. Le Gall SM, et al. (2010) ADAM17 is regulated by a rapid and reversible mechanismthat controls access to its catalytic site. J Cell Sci 123(Pt 22):3913–3922.

39. Soond SM, Everson B, Riches DW, Murphy G (2005) ERK-mediated phosphorylation ofThr735 in TNFalpha-converting enzyme and its potential role in TACE protein traf-ficking. J Cell Sci 118(Pt 11):2371–2380.

40. Xu P, Derynck R (2010) Direct activation of TACE-mediated ectodomain shedding byp38 MAP kinase regulates EGF receptor-dependent cell proliferation. Mol Cell 37(4):551–566.

41. Maxson JE, et al. (2014) Ligand independence of the T618I mutation in the colony-stimulating factor 3 receptor (CSF3R) protein results from loss of O-linked glycosyla-tion and increased receptor dimerization. J Biol Chem 289(9):5820–5827.

42. Cambien F, et al. (1992) Deletion polymorphism in the gene for angiotensin-con-verting enzyme is a potent risk factor for myocardial infarction. Nature 359(6396):641–644.

43. Keavney B, et al. (2000) Large-scale test of hypothesised associations between theangiotensin-converting-enzyme insertion/deletion polymorphism and myocardial in-farction in about 5000 cases and 6000 controls. International Studies of Infarct Sur-vival (ISIS) Collaborators. Lancet 355(9202):434–442.

44. Ota T, et al. (2004) Complete sequencing and characterization of 21,243 full-lengthhuman cDNAs. Nat Genet 36(1):40–45.

45. Hawkins GA, et al. (2012) The IL6R variation Asp(358)Ala is a potential modifier oflung function in subjects with asthma. J Allergy Clin Immunol 130(2):510–515.

46. Zheng Y, Saftig P, Hartmann D, Blobel C (2004) Evaluation of the contribution ofdifferent ADAMs to tumor necrosis factor alpha (TNFalpha) shedding and of thefunction of the TNFalpha ectodomain in ensuring selective stimulated shedding bythe TNFalpha convertase (TACE/ADAM17). J Biol Chem 279(41):42898–42906.

14628 | www.pnas.org/cgi/doi/10.1073/pnas.1511175112 Goth et al.