Dynamic regulation of FGF23 by Fam20Cphosphorylation, GalNAc-T3 glycosylation,and furin proteolysisVincent S. Tagliabraccia, James L. Engela,1, Sandra E. Wileya, Junyu Xiaoa, David J. Gonzaleza,b,Hitesh Nidumanda Appaiahc, Antonius Kollerd, Victor Nizetb,e, Kenneth E. Whitec, and Jack E. Dixona,f,g,2

Departments of aPharmacology, bPediatrics, fCellular and Molecular Medicine, and gChemistry and Biochemistry and eSkaggs School of Pharmacy andPharmaceutical Sciences, University of California, San Diego, La Jolla, CA 92093; cDepartment of Medical and Molecular Genetics, Indiana University School ofMedicine, Indianapolis, IN 46202; and dStony Brook University Proteomics Center, School of Medicine, Stony Brook University, Stony Brook, NY 11794

Contributed by Jack E. Dixon, February 8, 2014 (sent for review January 23, 2014)

The family with sequence similarity 20, member C (Fam20C) hasrecently been identified as the Golgi casein kinase. Fam20C phosphor-ylates secreted proteins on Ser-x-Glu/pSer motifs and loss-of-functionmutations in the kinase cause Raine syndrome, an often-fatal osteo-sclerotic bone dysplasia. Fam20C is potentially an upstream regulatorof the phosphate-regulating hormone fibroblast growth factor 23(FGF23), because humans with FAM20C mutations and Fam20C KOmice develop hypophosphatemia due to an increase in full-length, bi-ologically active FGF23. However, the mechanism by which Fam20Cregulates FGF23 is unknown. Here we show that Fam20C directlyphosphorylates FGF23 on Ser180, within the FGF23 R176XXR179/S180AEsubtilisin-like proprotein convertase motif. This phosphorylation eventinhibits O-glycosylation of FGF23 by polypeptide N-acetylgalactosami-nyltransferase 3 (GalNAc-T3), and promotes FGF23 cleavage and inac-tivation by the subtilisin-like proprotein convertase furin. Collectively,our results provide a molecular mechanism by which FGF23 is dynam-ically regulated by phosphorylation, glycosylation, and proteolysis.Furthermore, our findings suggest that cross-talk between phosphor-ylation and O-glycosylation of proteins in the secretory pathway maybe an important mechanism by which secreted proteins are regulated.

phosphate homeostasis | rickets | Fam20 | familial tumoral calcinosis |chronic kidney disease

Protein kinases are evolutionarily conserved enzymes thatregulate numerous cellular processes by transferring a mole-

cule of phosphate from ATP to target substrates (1, 2). The vastmajority of theses enzymes function within the nucleus and cy-tosol. In contrast, there are several examples of phosphorylatedproteins that are secreted from the cell, which raises the ques-tion: What are the kinases that phosphorylate these secretedphosphoproteins? We recently identified a small family of se-cretory pathway kinases that phosphorylate secreted proteinsand proteoglycans (3). These enzymes have N-terminal signalsequences that direct them to the lumen of the endoplasmicreticulum, where they encounter the proteins or proteoglycansthat they phosphorylate. One member of this atypical kinasefamily is the family with sequence similarity 20, member C(Fam20C), which phosphorylates secreted proteins on Ser(S)-x-Glu(E)/pSer(pS) (S-x-E/pS) motifs (3, 4). Because Fam20Clocalizes within the secretory pathway and the vast majority ofsecreted phosphoproteins are phosphorylated on S-x-E/pSmotifs, Fam20C has been proposed to play a major role in thegeneration of the secreted phosphoproteome (5–7). For exam-ple, ∼75% of human serum and cerebrospinal fluid phospho-proteins are phosphorylated on S-x-E/pS motifs (8, 9). Thisincludes proteins important for tooth and bone formation, aswell as numerous hormones. In most cases, the functional im-portance of these phosphorylation events is unknown.Loss-of-function mutations in the human FAM20C gene cause

Raine syndrome, an often-fatal osteosclerotic bone dysplasia(10, 11). Most Raine patients die within the first few weeks of

life. Nonlethal cases have also been reported, and these patientsdevelop hypophosphatemia as a result of elevated levels of thephosphate-regulating hormone fibroblast growth factor 23 (FGF23)(12). Additionally, Fam20C knockout (KO) mice develop renalphosphate wasting due to an increase in circulating bioactiveFGF23 as well as severe hypophosphatemic rickets (13, 14).Thus, Fam20C has been proposed to be a regulator of FGF23;however, the molecular mechanisms underlying the control ofthis hormone by Fam20C are unclear.FGF23 is secreted from osteoblasts and osteocytes, and targets

the kidney to regulate the reabsorption of phosphate and ca-tabolism of 1,25-dihydroxyvitamin D3 (15, 16). FGF23 inhibitsrenal phosphate transport by activating FGF receptors (FGFRs)in a manner that requires binding the coreceptor α-klotho (α-KL)(17). Binding of FGF23 to FGFRs/α-KL induces urinary excretionof phosphate by decreasing the abundance of the type II sodium-dependent phosphate cotransporters NPT2a and NPT2c (15, 16).Several genetic disorders of renal phosphate wasting are associatedwith alterations in FGF23 (18). X-linked hypophosphatemia is themost prevalent form of hypophosphatemic rickets (1:20,000), and iscaused by loss-of-function mutations in the gene encoding thephosphate-regulating endopeptidase homolog X-linked, which isassociated with elevated FGF23 expression (19, 20). A similarsituation exists in patients with an autosomal recessive form ofhypophosphatemic rickets caused by mutations in the ectonucleotidepyrophosphatase ENPP1 (21) and the Fam20C substrate dentin

Significance

The family with sequence similarity 20, member C (Fam20C) isa secretory pathway-specific kinase that phosphorylates se-creted proteins on Ser-x-Glu/pSer motifs. Mutations in humanFAM20C cause a devastating childhood disorder known asRaine syndrome. Some patients with FAM20C mutations aswell as Fam20C KO mice develop hypophosphatemia due toelevated levels of the phosphate-regulating hormone FGF23.In this paper, we show that Fam20C phosphorylates FGF23 ona Ser-x-Glu motif that lies within a critical region of the hor-mone. The phosphorylation promotes FGF23 proteolysis by furinby blocking O-glycosylation by polypeptide N-acetylgalactosami-nyltransferase 3. Our results have important implications forpatients with abnormalities in phosphate homeostasis.

Author contributions: V.S.T., J.L.E., S.E.W., J.X., and J.E.D. designed research; V.S.T., J.L.E.,S.E.W., J.X., D.J.G., and A.K. performed research; H.N.A., A.K., V.N., and K.E.W. contrib-uted new reagents/analytic tools; V.S.T., J.L.E., S.E.W., J.X., D.J.G., H.N.A., A.K., V.N., K.E.W.,and J.E.D. analyzed data; and V.S.T., K.E.W., and J.E.D. wrote the paper.

The authors declare no conflict of interest.

Freely available online through the PNAS open access option.1Present address: Molecular, Cellular & Integrative Physiology Graduate Program, Depart-ment of Medicine, University of California, Los Angeles, CA 90095.

2To whom correspondence should be addressed. E-mail: jedixon@ucsd.edu.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1402218111/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1402218111 PNAS Early Edition | 1 of 6

BIOCH

EMISTR

Y

matrix protein 1 (DMP1) (22). These disorders share the com-mon denominator of elevated FGF23.Other Mendelian disorders of renal phosphate handling have

provided important insight into the regulation of FGF23 proteinprocessing. FGF23 is inactivated in the Golgi by proteolysis withina highly conserved subtilisin-like proprotein convertase (SPC) site,176RHTR179/S180AE182, generating inactive N- and C-terminalfragments (23). Gain-of-function missense mutations in FGF23cause autosomal dominant hypophosphatemic rickets (ADHR)(24). These mutations substitute the Arg (R) residues within theSPC cleavage site and render the protein resistant to proteolysis(25). Conversely, loss-of-function mutations in FGF23 cause fa-milial tumoral calcinosis (FTC), a hyperphosphatemic disordercharacterized by often severe ectopic and vascular calcifications(26). FTC is also caused by loss-of-function mutations inpolypeptide N-acetylgalactosaminyltransferase 3 (GalNAc-T3), a glycosyltransferase that O-glycosylates FGF23 atThr178 within the SPC cleavage site (27, 28). These inactivatingGalNAc-T3 mutations prevent O-glycosylation of FGF23, whichproduces a hormone more susceptible to SPC proteolysis (28,29). Collectively, these studies suggest that FGF23 processing isa highly regulated and physiologically important process.Here we demonstrate that Fam20C regulates FGF23 by

phosphorylation of Ser180, a residue that neighbors the SPCsite (R176H177T178R179/S180). We show that Ser180 phosphory-lation inhibits GalNAc-T3 O-glycosylation of Thr178, thereby pro-moting furin (PCSK3)-dependent FGF23 proteolysis. Our resultsprovide a plausible molecular mechanism by which loss of

Fam20C leads to elevated levels of intact, biologically active FGF23and subsequent hypophosphatemia. Furthermore, our results sug-gest that cross-talk between phosphorylation and O-glycosylation ofproteins in the secretory pathway may be an important mechanismby which secreted proteins are dynamically regulated.

ResultsFGF23 Is Phosphorylated by Fam20C in Vitro and in Cells. Contem-poraneous with our finding that Fam20C is a protein kinase, Wanget al. (13) characterized Fam20C KO mice and demonstrated thatthese animals develop hypophosphatemic rickets as a result ofelevated serum intact, bioactive FGF23. Furthermore, a recentreport identified compound heterozygous mutations in FAM20Cin two siblings referred for hypophosphatemia and severe dentaldemineralization disease (12). Biochemical analysis of the patients’sera identified elevated intact FGF23. Because FGF23 is expressedin tissues that show high levels of Fam20C expression (30) andcontains Fam20C consensus S-x-E/pS motifs (Fig. 1A), we hy-pothesized that Fam20C may directly regulate FGF23 by phos-phorylation. To determine whether FGF23 is a phosphoprotein,we generated a HEK293T cell line stably expressing a protease-resistant mutant of FGF23 (FGF23 R176Q) (25). This con-struct contained a C-terminal Flag tag that allowed us toaffinity-purify it from conditioned medium. We then digestedFGF23 R176Q with trypsin or chymotrypsin, separated thepeptides by liquid chromatography, and analyzed the pep-tides by mass spectrometry (MS). Tandem mass spectrometry(MS/MS) analysis identified 180pSAEDDSERDPLNVLKPR196

and 178TRpSAEDDSERDPL190 in the trypsin and chymotrypsindigests, respectively (Fig. 1 B and C and Fig. S1). These resultssuggest that FGF23 is phosphorylated in HEK293T cells on Ser180.Notably, Ser180 lies within a Fam20C S-x-E recognition motif andneighbors the SPC recognition site (Fig. 1A). To test whetherFam20C phosphorylates FGF23, we performed in vitro kinasereactions with recombinant Fam20C and FGF23 R176Q. Fam20Cphosphorylated FGF23 R176Q in a time-dependent manner,whereas the catalytically inactive Fam20C D478A failed to in-corporate phosphate (Fig. 1D). Furthermore, treatment of pu-rified FGF23 R176Q with recombinant Fam20C increased therelative abundance of Ser180 phosphorylation (Fig. S2). By cal-culating the area under the selected ion species, we estimatedthat treatment with Fam20C increased the relative abundance ofFGF23 R176Q Ser180 phosphorylation ∼10-fold.To determine whether Fam20C phosphorylates FGF23 in

cells, we used clustered regularly interspaced short palindromicrepeats (CRISPR/Cas9) genome-editing technology to delete theFAM20C gene from the osteoblast-like cell line U2OS (31, 32)(Fig. S3A). We generated U6-driven expression cassettes to ex-press single-guide RNAs targeting exon 1 of the human FAM20Cgene for expression in U2OS cells (Fig. S3 B and D). Two cloneswere detected to have insertions/deletions (indels) that resultedin frameshift mutations resulting in premature stop codons (Fig.S3 C and E). Protein immunoblotting with a polyclonal antibodyraised against full-length Fam20C detected Fam20C protein inthe conditioned medium of native U2OS cells but not the cellsthat were subjected to genome editing (Fig. 2A). To confirm lossof Fam20C activity, the phosphorylation status of the Fam20Csubstrate osteopontin (OPN) was analyzed in control andFam20C KO cells that were metabolically labeled with [32P]ortho-phosphate. V5-tagged OPN was immunoprecipitated from con-ditioned medium, and OPN protein and phosphorylation levelswere analyzed by autoradiography. OPN was phosphorylated incontrol cells, and no detectable phosphorylation was observedin Fam20C KO cells (Fig. 2B).In U2OS cells, ectopically expressed, C-terminal V5-tagged

FGF23 is O-glycosylated, and we can detect the full-lengthhormone and the C-terminal fragments in the conditionedmedium by protein immunoblotting of V5-immunoprecipitates(Fig. 2C). We transiently transfected V5-tagged FGF23 in con-trol and Fam20C KO cells that we metabolically labeled with[32P]orthophosphate and analyzed V5-immunoprecipitates from

A

300 400 500 600 700 800 900 1000 1200 1300 1400 1500 1600 1700 1800 19000

20

40

60

80

100 936.5435

712.6581 1005.3077

770.1674

604.6067827.6633

488.2676 919.5739 1051.5223 1669.58831541.49511329.3741400.3679 513.4383 1428.43561152.5686 1784.6434

300 400 500 600 700 800 900 1000 1100 1200 1300 1400 1500 1600 1700 1800 19000

5

10

15

20

25

Rel

ativ

e A

bund

ance

962.1870

936.5340770.1452712.6048

1085.2113695.0972 827.6276604.5301437.3246 1164.4706 1749.4931547.3577322.2546 1651.57571279.0575 1410.4676 1473.0205 1847.4840

1100

1865.6

+2

+2

+2

+2

+2

+2 +2

(MH-H3PO4)+2

+2

180S A E D D S E R D P L N V L K P R196

180Sp A E D D S E R D P L N V L K P R196

y14 y13 y12 y10 y9 y8

b9 b12 b13 b14 b15

y14 y12y13 y10 y9 y8

b9 b12 b15

+phos+phos +phos

SP cFGF23nFGF23

176R-H-T-R S-A-E182

1 24 251SPC site Coomassie

FGF23 R176Q

Autorad

(min) 0 1 5 10 30 60 120 120DA

B

C

D

iFGF23

176Q/W-H-T-Q/W-S-A-E182ADHR

Mutations

y8

y10

y12

y13

y14

b9

y9 b12 b13b14 b15

m/z

y10

y12y13

y14

y8

b9 b12b15

Fig. 1. Fam20C phosphorylates FGF23 on Ser180. (A) Schematic representationof human FGF23 indicating the signal peptide (SP), intact FGF23 (iFGF23), N-terminal fragment (nFGF23), and C-terminal fragment (cFGF23). The residuessurrounding the SPC cleavage site (downward arrow) are shown. Mutationsthat replace the Arg in patients with autosomal dominant hypophosphatemicrickets (ADHR) are also shown. A potential Fam20C phosphorylation site is inred. (B and C) Representative MS/MS fragmentation spectra of a tryptic peptide(FGF23 180–196) depicting Ser180 phosphorylation of FGF23 R176Q purifiedfrom conditioned medium of HEK293T cells. (D) Time-dependent incorporationof 32P from [γ-32P]ATP into FGF23 R176Q by Fam20C or Fam20C D478A (DA).Reaction products were analyzed by SDS/PAGE and autoradiography.

2 of 6 | www.pnas.org/cgi/doi/10.1073/pnas.1402218111 Tagliabracci et al.

conditioned medium. FGF23 was phosphorylated in control butnot in Fam20C-deficient cells, and phosphorylation was moreprevalent within the C-terminal fragment (Fig. 2D). Thus,Fam20C phosphorylates FGF23 in vitro and in cells.

Phosphorylation of FGF23 at Ser180 Prevents O-Glycosylation by GalNAc-T3.The physiological significance of FGF23 O-glycosylation wasunderscored when inactivating mutations in the gene encod-ing GalNAc-T3 (GALNT3) were found in patients with FTC(27). Patients with FTC develop hyperphosphatemia and se-vere ectopic calcifications, symptoms of which are the metabolic“mirror image” of the hypophosphatemic disorders (26). Sub-sequent studies identified inactivating mutations in FGF23 inFTC patients without mutations in GALNT3, suggesting thatFGF23 and GalNAc-T3 operate in a common pathway to reg-ulate phosphate homeostasis (26). The polypeptide N-acetylga-lactosaminyltransferase (GalNAc-transferase) family consists of20 enzymes that catalyze the initial step of mucin-type O-glycosyl-ation by transferring GalNAc from UDP-GalNAc to Ser and Thrresidues in secretory pathway proteins (33). GalNAc-T3 O-glyco-sylates FGF23 on Thr178 within the R176H177T178R179/S180 SPC site,and has been shown to protect FGF23 from proteolytic cleavageand thus inactivation (28). Because Fam20C phosphorylates FGF23at Ser180, and this residue is located immediately adjacent tothe SPC site (Fig. 1A), we hypothesized that phosphorylation mayregulate FGF23 O-glycosylation and/or proteolytic processing. InFam20C-deficient U2OS cells, we did not observe changes inO-glycosylation or proteolytic processing of ectopically expressedFGF23. However, it is likely that endogenous Fam20C levels and/or activity are much lower than the glycosyltransferases and/orproteases that modify FGF23 in these cells (i.e., Fam20C is lim-iting with respect to the glycosyltransferases and/or proteases).Deletion of Fam20C therefore would have no effect on O-glyco-sylation and/or processing. To explore the effect of phosphoryla-tion of FGF23 on O-glycosylation and proteolytic processing, wecoexpressed Flag-tagged Fam20C or the catalytically inactiveFam20C D478A mutant with V5-tagged FGF23 in U2OS cellsand analyzed immunoprecipitates from conditioned medium.Wild-type Fam20C, but not the inactive D478A mutant, decreasedthe mobility of the C-terminal fragments of FGF23 that was re-versed by λ-phosphatase treatment, suggesting a phosphorylationevent (Fig. 3A, Lower, first, second, and third lanes). Interestingly,the doublet in full-length, intact FGF23 (iFGF23) was absentwhen active Fam20C, but not the inactive Fam20C D478A mutant,was expressed (Fig. 3A, Upper, first and second lanes). Given theclose proximity of Ser180 to the SPC site, we reasoned that phos-phorylation of FGF23 at Ser180 would prevent O-glycosylation atThr178 and therefore explain the loss of the doublet observed iniFGF23 induced by Fam20C (Fig. 3A, second lane). Indeed, muta-tion of Ser180 to Ala prevented the loss of the doublet in iFGF23when Fam20C was expressed (Fig. 3A, fourth lane). Consistently,mutation of Thr178 to Ala (a nonglycosylated mutant) or Ser180 toAsp (a phosphomimetic) resulted in a species that migrated similarto the phosphorylated WT iFGF23 (Fig. 3A, fifth and sixth lanes).These results support that Fam20C phosphorylation of FGF23 atSer180 prevents O-glycosylation at Thr178.To determine whether phosphorylation of FGF23 Ser180 di-

rectly affects O-glycosylation of FGF23 at Thr178 by GalNAc-T3,we produced a HEK293T cell line stably expressing a secretedform of GalNAc-T3 lacking the first 37 residues encompassingthe transmembrane domain and immunopurified the enzymefrom conditioned medium (sGalNAc-T3; residues 38–633; Fig.S4). We then performed in vitro glycosylation experiments withsGalNAc-T3 and a peptide substrate surrounding the SPCcleavage site of FGF23 [residues 172–190 of human FGF23;hereafter referred to as FGF23(172–190)]. MALDI-TOF MSanalysis of the reaction products demonstrated that sGalNAc-T3catalyzed the transfer of GalNAc from UDP-GalNAc to Thr178of FGF23(172–190) as expected (Fig. 3B and Figs. S5A and S6).However, when sGalNAc-T3 was assayed against Ser180-phosphor-ylated FGF23(172–190), no detectable glycosylation was

observed upon extended incubation (Fig. 3C and Fig. S5B). Col-lectively, our data support that Fam20C phosphorylatesFGF23 at Ser180 and inhibits O-glycosylation by GalNAc-T3 atThr178. Importantly, these results provide a molecular mecha-nism that could account for the increase in serum full-lengthFGF23 in Fam20C KO mice and in humans with FAM20Cmutations, namely reduced Fam20C phosphorylation allowing O-glycosylation and stabilization of FGF23.

Phosphorylated FGF23 Is Cleaved by the SPC Furin. The importanceof FGF23 cleavage is underscored in patients with ADHR whohave mutations that substitute the Arg residues within the SPCcleavage site and render the protein resistant to proteolysis(Fig. 1A) (24, 25). As our above results support, phosphorylationof FGF23 at Ser180 inhibits O-glycosylation and would thereforepromote hormone proteolysis and thus inactivation. The specificprotease that inactivates FGF23 has yet to be conclusivelyidentified, but is likely a member of the SPC family based upon

WT KO KO

15 KDa

25 KDa

35 KDa

15 KDa

25 KDa

35 KDa

B

70KDa

32P(OPN)

α-Fam20C

IP: V5(OPN)

250 KDa

130 KDa

100 KDa

70 KDa

55 KDa

35 KDa

α-Fam20C

C

α-GAPDH35 KDa

WT KO KO

25 KDa

35 KDa

15 KDa

35 KDa

O-glycosidase

α-Neuraminidase

PNGase F

++

+A

D

α-Fam20C

Medium

Medium

Extracts

Medium

glycosylated iFGF23non-glycosylated

iFGF23

cFGF23

IP: V5IB: V5

α-GAPDH

Extracts

IP: V5IB: V5

iFGF23

cFGF23

Medium

iFGF23

cFGF23

Medium

IP: V532P

C5

C9

Extracts

WT KO KO

C5

C9

C5

C9

Fig. 2. Fam20C phosphorylates FGF23 in mammalian cells (A) Protein im-munoblotting of trichloroacetic acid precipitates from conditioned mediumof control and Fam20C KO cells (C5, clone 5; C9, clone 9) using an affinity-purified rabbit anti-Fam20C polyclonal antibody. GAPDH from cell extracts isshown as loading control (Lower). (B) Control and Fam20C KO cells weremetabolically labeled with 32PO4

3− and transfected with V5-tagged OPN.Protein immunoblotting of V5-immunoprecipitates from conditioned me-dium (Top) and autoradiography depicting 32P incorporation into OPN(Middle). Fam20C protein levels in conditioned medium are also shown(Bottom). (C) Protein immunoblotting of V5-immunoprecipitates from theconditioned medium of U2OS cells expressing V5-tagged FGF23. V5-immuno-precipitates were treated with O-glycosidase and α-(2→3,6,8,9)-neuraminidaseto remove O-linked glycosylation or PNGase F to remove N-linked glycosyla-tion. (D) Control and Fam20C KO cells were metabolically labeled with 32PO4

3−

and transfected with V5-tagged FGF23. Protein immunoblotting of V5-immunoprecipitates from conditioned medium (Top) and autoradiographydepicting 32P incorporation into FGF23 (Middle). Intact FGF23 and C-terminalfragments are shown. Fam20C protein levels in conditioned medium are alsoshown (Bottom). IB, immunoblotting; IP, immunoprecipitation.

Tagliabracci et al. PNAS Early Edition | 3 of 6

BIOCH

EMISTR

Y

the RXXR sequence comprising the cleavage site. With re-spect to our model, the protease must cleave FGF23 betweenArg179 and a phosphorylated Ser180 (176RHTR↓pSAE182). Toexplore the proteolytic processing of phosphorylated FGF23,we coexpressed Flag-tagged PCSK1, PCSK2, and PCSK3 (furin)with V5-tagged FGF23 in U2OS cells. Coexpression of theproteases and FGF23 had no detectable effect on the relativelevels of full-length secreted WT FGF23 (Fig. 4A). However,when FGF23 S180D (a phosphomimetic residue) or T178A (anO-glycosylation–defective mutant) was coexpressed with the pro-teases, the SPC furin completely abolished the iFGF23 but didnot concurrently increase the levels of the C-terminal fragments(Fig. 4A). We reasoned that this could potentially be due to thepresence of additional furin-specific RXXR recognition motifswithin the C terminus of FGF23 that are cleaved by overex-pressed furin. Therefore, to resolve this experimental possibility,we deleted furin in U2OS cells using CRISPR/Cas9 genomeediting (Fig. S7). When WT FGF23 was expressed in furin KOcells, only intact FGF23 was detected, suggesting that furincleaves FGF23 (Fig. 4B). Moreover, we analyzed the ability ofrecombinant furin to cleave the Ser180-phosphorylated FGF23(172–190) peptide by MALDI-TOF MS. Incubation of the non-phosphorylated or phosphorylated peptide with furin resultedin the time-dependent cleavage of both substrates (Fig. 4 C and Dand Fig. S8). Thus, phosphorylation of FGF23 at Ser180 preventsThr178 O-glycosylation by GalNAc-T3, which then allows furin-dependent proteolysis.

Incomplete Inhibition of FGF23 O-Glycosylation by Fam20C T268M.Raine syndrome was originally described in 1989 as an aggres-sive, neonatal osteosclerotic bone dysplasia that results in deathwithin the first few weeks of life (11). Subsequently, mutations inthe FAM20C gene were found to be responsible for this disorder(10), and recent reports suggest a broader phenotypic spectrumfor individuals with FAM20C mutations because some patientssurvive infancy (12, 34). Notably, Rafaelsen et al. (12) identifieda novel missense mutation in Fam20C that substituted Thr268

with a Met in two compound heterozygous brothers who hadelevated serum intact FGF23 and hypophosphatemia. Threonine268 (Thr175 in Caenorhabditis elegans) is highly conserved withinthe Fam20C subfamily, and the C. elegans Fam20 crystal struc-ture revealed that this Thr is located in the glycine-rich loop(β1-β2), a region important for nucleotide binding (35) (Fig. 5 Aand B). To test the impact of this mutation on FGF23 phos-phorylation and O-glycosylation, we generated Flag-taggedFam20C with Thr268 mutated to Met (Fam20C T268M) and ana-lyzed its activity as well as the ability to be secreted from U2OScells. In contrast to other nonlethal Fam20C mutations (3), sub-stitution of Thr268 with Met did not affect Fam20C secretion (Fig. 5C and D). Fam20C T268M phosphorylated OPN and FGF23much less efficiently than WT Fam20C, as judged by its ability

to induce a mobility change in OPN (Fig. 5C) and the secreted,C-terminal fragment of FGF23 (Fig. 5D). Importantly, expressionof Fam20C T268M did not completely prevent O-glycosylation ofintact FGF23, whereas WT Fam20C achieved complete inhibitionof FGF23 O-glycosylation (Fig. 5D). Taken together, these findingssupport that the elevated intact FGF23 and hypophosphatemiaobserved in patients with the Fam20C T268Mmutation are a resultof the inability of Fam20C to efficiently phosphorylate FGF23to inhibit O-glycosylation. As a consequence, elevated biologicallyactive, O-linked glycosylated FGF23 is secreted, ultimately resultingin hypophosphatemia.

DiscussionThe results of this study expand our understanding of the com-plex control of mammalian phosphate homeostasis. Our modelfor the regulation of FGF23 envisions a highly dynamic interplaybetween O-glycosylation by GalNAc-T3 (or other members ofthe GalNAc-transferase family) and phosphorylation by Fam20Cas a means by which to balance the processing of FGF23 as in-tact, biologically active protein (glycosylated > phosphorylated)or N- and C-terminal fragments (phosphorylated > glycosylated)(Fig. S9). This model is consistent with the marked elevation ofintact, biologically active FGF23 observed in humans with theFam20C T268M mutation (12) and in Fam20C KO mice (13, 14).Sensing the need to adapt phosphate balance in vivo is a

complex process, and our data support that the production of bio-active FGF23 will depend upon, in addition to the levels of FGF23transcription, the relative expression and activities of osteoblast/osteocyte Fam20C, GalNAc-T3, and furin. The mechanismsby which Fam20C is functionally or expressionally regulated areunknown. Recent studies examining the production of intact and/or C-terminal fragments of FGF23 in fibrous dysplasia and duringphysiological situations of low iron, such as in ADHR (36, 37),support that GalNAc-T3 and furin are likely controlled bymultiple factors including 3′,5′-cyclic adenosine monophosphate(38), iron or iron deficiency (39), and inorganic phosphate (40).Site-specific O-glycosylation within, or neighboring, SPC sites

is emerging as an important regulatory mechanism to controlSPC-dependent processing of secreted proteins (41). Our resultsare in accord with the concept that phosphorylation of hormonesmay promote proteolysis by SPCs (or other proteases) by in-terfering with O-glycosylation by members of the GalNAc-transferase family. Our observation of competition betweenphosphorylation and O-glycosylation in the secretory pathwayreflects prior observations on the cross-talk between phosphor-ylation and O-glycosylation by O-linked N-acetylglucosamine(O-GlcNAc) transferase, which transfers GlcNAc from UDP-GlcNAc to Ser and Thr residues in nuclear and cytosolicproteins (42). Interplay between O-glycosylation and phosphor-ylation of the tumor suppressor p53 has been shown to co-ordinately regulate p53 stability and activity (43). Analogous

PIPRRHTRSAEDDSERDPL FGF23(172-190)

20C

D47

8A20

C W

T 20

C W

T +

λ p’

tase

20C

WT

20C

D47

8A

FGF23

A

20C

D47

8A

WT S180AT178A S180D

α-Flag (Fam20C)

Medium

Extracts

B

15 KDa

25 KDa

35 KDa

35 KDa

70 KDa

35 KDa

glycosylated iFGF23non-glycosylated

iFGF23

C

PIPRRHTRpSAEDDSERDPL pSer180 FGF23(172-190)

cFGF23

α-GAPDH

IP: V5IB: V5

α-V5 (iFGF23)

Fig. 3. Phosphorylation of FGF23 at Ser180 inhibitsO-glycosylation by GalNAc-T3. (A) Protein immuno-blotting of V5-immunoprecipitates from the condi-tioned medium of U2OS cells expressing V5-taggedFGF23 or mutants (S180A, T178A, S180D) with eitherWT (20C) or catalytically inactive D478A (20C D478A)FLAG-tagged Fam20C. V5-immunoprecipitates weretreated with λ-phosphatase (λp’tase) (Upper). Extractswere analyzed for Fam20C, Fam20C D478A, FGF23,and GAPDH (Lower). (B and C ) sGalNAc-T3 wasincubated overnight (o/n) with FGF23(172–190)(B) or Ser180-phosphorylated FGF23(172–190) (C )and UDP-GalNAc. The products were analyzed byMALDI-TOF MS.

4 of 6 | www.pnas.org/cgi/doi/10.1073/pnas.1402218111 Tagliabracci et al.

to what we observed in this work, phosphorylation of p53promotes its degradation by the ubiquitin system by inhibitingO-GlcNAcylation. Currently, it is unknown whether the interplaybetween O-glycosylation by members of the GalNAc-transferasefamily and phosphorylation by secretory pathway kinases isa common mechanism within the secretory pathway or whetherFGF23 is a unique example. However, it is worth noting thatbone morphogenetic protein 15 (BMP15), a secreted maternalhormone essential for mammalian reproduction, is phosphory-lated by Fam20C on an S-x-E motif (44). This phosphorylationsite (Ser6) is located close to an O-glycosylated Thr (Thr10).When BMP15 is expressed in HEK293 cells, phosphorylation andO-glycosylation appear to be mutually exclusive (45).The physiological scenario for the regulated control of circu-

lating FGF23 is likely far more complicated than direct in-teractions between FGF23, Fam20C, GalNAc-T3, and furin. Itis possible that Fam20C regulates FGF23 not only through direct

mechanisms, as our results suggest, but also through emerging,indirect pathways. In this regard, we have shown that DMP1,a highly phosphorylated extracellular matrix protein critical forproper mineralization of bone (46), is a substrate for Fam20C (3,4). Inactivating mutations in DMP1 result in autosomal recessivehypophosphatemic rickets, and Dmp1 KO mice share manyphenotypic similarities with those of Fam20C-deficient animals,including elevated FGF23 (13, 22). DMP1 appears to regulateFGF23 expression indirectly, as loss of Dmp1 in vivo impairs thematuration of osteoblasts to osteocytes through unknown mech-anisms and results in highly elevated FGF23 mRNA as well ascirculating protein (22). Thus, loss of DMP1 phosphorylation ina state of Fam20C deficiency could also contribute to an increasein FGF23 in vivo. Certainly additional studies will be required tounderstand these new interactions.In addition to Ser180, three Fam20C consensus S-x-E/pS sites

are present in the C-terminal fragment of FGF23 (residues 180–250). We were unable to detect FGF23 peptides surrounding thepredicted sites by MS, and therefore at this time we cannot ruleout Fam20C-dependent phosphorylation of these residues. Byquantitating the incorporated radioactivity in Fig. 1D, we calcu-lated a stoichiometry of about 1.5–2 mol of phosphate per molFGF23 R176Q, suggesting that other sites may be phosphorylated.In conclusion, we have demonstrated that Fam20C regulates

FGF23 by phosphorylation of Ser180. Phosphorylation of FGF23inhibits GalNAc-T3 O-glycosylation, which then allows proteolysisby furin. Collectively, our results provide a mechanism by whichloss of Fam20C leads to aberrations in phosphate homeostasis dueto a cellular shift toward increased O-glycosylation and thus ele-vated levels of intact, biologically active FGF23 and, conversely,increased Fam20C activity shifts the FGF23 processing balancetoward increased furin proteolysis and hormone inactivation.Furthermore, our data suggest that interplay between phos-phorylation and O-glycosylation of proteins in the secretory

α-GAPDH

IP: V5IB: V5

PC

SK

1P

CS

K2

Furin

PC

SK

1P

CS

K2

Furin

PC

SK

1P

CS

K2

Furin

iFGF23

cFGF23

α-Flag (Proteases)

α-V5 (iFGF23)

FGF23 S180D T178A

Medium

Extracts

A

100 KDa

25 KDa

35 KDa

35 KDa

15 KDa

70 KDa

35 KDa

B

α-GAPDH

IP: V5IB: V5

iFGF23

cFGF23

α-V5 (iFGF23)

Medium

Extracts

FURIN WT KO KO

25 KDa

35 KDa

15 KDa

35 KDa

35 KDa

α-Furin

WT KO KO

C17

C82

C17

C82

100 KDa

WT

C D

Fig. 4. Furin cleaves phosphorylated FGF23. (A) Protein immunoblottingof V5-immunoprecipitates from the conditioned medium of U2OS cells ex-pressing V5-tagged FGF23 or mutants (S180D and T178A) with FLAG-taggedPCSK1, PCSK2, or PCSK3 (furin) (Upper). Extracts were analyzed for PCSK1–3,FGF23, and GAPDH expression (Lower). (B) Protein immunoblotting of V5-immunoprecipitates from the conditioned medium of control or FURIN KOU2OS cells (C17, clone 17; C82, clone 82) (Upper). Extracts were analyzed forFGF23 and GAPDH expression (Lower). (C and D) Time-dependent cleavageof FGF23(172–190) (C) or Ser180-phosphorylated FGF23(172–190) (D). Theproducts were analyzed by MALDI-TOF MS.

20C

D47

8A20

C W

T 20

C T

268M

70 KDa

35 KDa

35 KDa

15 KDa

70 KDa

25 KDa

Thr175

A BH.sapiens MKSGGTQLKM.musculus MKSGGTQLKR.norvegicus MKSGGTQLKD.melanogaster QKEGGTQLKC.elegans IMDGGTQVK

D

glycine

rich loop

(OPN)

70 KDa

70 KDa

70 KDa

100 KDa

35 KDa

20C

D47

8A20

C W

T 20

C T

268MC

IP: FlagIB: Flag

(Fam20C)

α-Flag (Fam20C)

Medium

Extracts

glycosylated iFGF23non-glycosylated

iFGF23

cFGF23

α-GAPDH

IP: V5IB: V5

Medium

IP: V5IB: V5

Medium IP: Flag

IB: Flag(Fam20C)

α-Flag (Fam20C)

Extracts α-GAPDH

Fig. 5. Deficient inhibition of FGF23 O-glycosylation by Fam20C T268M. (A)Structural representation of the active site of Fam20 from C. elegans high-lighting the position of Thr175 (human Thr268). The glycine-rich loop and nu-cleotide (ADP) are shown. (B) Sequence alignment of Fam20C from human(Homo sapiens), mouse (Mus musculus), rat (Rattus norvegicus), fly (Drosophilamelanogaster), and worm (Caenorhabditis elegans) depicting the conservationof the Thr. (C and D) Protein immunoblotting of V5 and Flag-immunopreci-pitates from conditioned medium of U2OS cells expressing V5-tagged OPN (C)or FGF23 (D) with Flag-tagged D478A (20C D478A), WT (20C WT), or mutantFam20C found in patients with FGF23-related hypophosphatemia (20C T268M)(Upper). Extracts were analyzed for Fam20C and GAPDH expression (Lower).

Tagliabracci et al. PNAS Early Edition | 5 of 6

BIOCH

EMISTR

Y

pathway may be a critical posttranslational mechanism by whichsecreted proteins are regulated.

MethodsProtein Purification. Flag-tagged Fam20C and FGF23 R176Q were immuno-purified from conditioned medium of HEK293T cells as previously described(3). More details on protein purification are presented in SI Methods.

Generation of Anti-Fam20C Antibody. Polyclonal antisera were raised in rab-bits against recombinant Flag-tagged Fam20C produced in HEK293T cells

(Cocalico Biologicals). More details on antibody generation are presented inSI Methods.

ACKNOWLEDGMENTS. We thank Carolyn A. Worby, Gregory S. Taylor, Jenna L.Jewell, and members of the J.E.D. laboratory for insightful discussionsand comments regarding the manuscript. We thank David King of theHoward Hughes Medical Institute’s Mass Spectrometry Laboratory (Uni-versity of California, Berkeley) for peptide synthesis and Eric Durrant fortechnical assistance. This work was supported in part by National Institutes ofHealth Grants DK018849-36 and DK018024-37 (to J.E.D.), DK63934 (to K.E.W.),and K99DK099254 (to V.S.T.).

1. Cohen P (2002) The origins of protein phosphorylation. Nat Cell Biol 4(5):E127–E130.2. Fischer EH (2013) Cellular regulation by protein phosphorylation. Biochem Biophys

Res Commun 430(2):865–867.3. Tagliabracci VS, et al. (2012) Secreted kinase phosphorylates extracellular proteins

that regulate biomineralization. Science 336(6085):1150–1153.4. Ishikawa HO, Xu A, Ogura E, Manning G, Irvine KD (2012) The Raine syndrome protein

FAM20C is a Golgi kinase that phosphorylates bio-mineralization proteins. PLoS ONE7(8):e42988.

5. Tagliabracci VS, Pinna LA, Dixon JE (2013) Secreted protein kinases. Trends BiochemSci 38(3):121–130.

6. Tagliabracci VS, Xiao J, Dixon JE (2013) Phosphorylation of substrates destined forsecretion by the Fam20 kinases. Biochem Soc Trans 41(4):1061–1065.

7. Salvi M, Cesaro L, Tibaldi E, Pinna LA (2010) Motif analysis of phosphosites disclosesa potential prominent role of the Golgi casein kinase (GCK) in the generation ofhuman plasma phospho-proteome. J Proteome Res 9(6):3335–3338.

8. Zhou W, et al. (2009) An initial characterization of the serum phosphoproteome.J Proteome Res 8(12):5523–5531.

9. Bahl JM, Jensen SS, Larsen MR, Heegaard NH (2008) Characterization of the humancerebrospinal fluid phosphoproteome by titanium dioxide affinity chromatographyand mass spectrometry. Anal Chem 80(16):6308–6316.

10. Simpson MA, et al. (2007) Mutations in FAM20C are associated with lethal osteo-sclerotic bone dysplasia (Raine syndrome), highlighting a crucial molecule in bonedevelopment. Am J Hum Genet 81(5):906–912.

11. Raine J, Winter RM, Davey A, Tucker SM (1989) Unknown syndrome: Microcephaly,hypoplastic nose, exophthalmos, gum hyperplasia, cleft palate, low set ears, andosteosclerosis. J Med Genet 26(12):786–788.

12. Rafaelsen SH, et al. (2013) Exome sequencing reveals FAM20c mutations associatedwith FGF23-related hypophosphatemia, dental anomalies and ectopic calcification.J Bone Miner Res 28(6):1378–1385.

13. Wang X, et al. (2012) Inactivation of a novel FGF23 regulator, FAM20C, leads to hy-pophosphatemic rickets in mice. PLoS Genet 8(5):e1002708.

14. Vogel P, et al. (2012) Amelogenesis imperfecta and other biomineralization defects inFam20a and Fam20c null mice. Vet Pathol 49(6):998–1017.

15. Bhattacharyya N, Chong WH, Gafni RI, Collins MT (2012) Fibroblast growth factor 23:State of the field and future directions. Trends Endocrinol Metab 23(12):610–618.

16. Farrow EG, White KE (2010) Recent advances in renal phosphate handling. Nat RevNephrol 6(4):207–217.

17. Kurosu H, et al. (2006) Regulation of fibroblast growth factor-23 signaling by klotho.J Biol Chem 281(10):6120–6123.

18. Carpenter TO (2012) The expanding family of hypophosphatemic syndromes. J BoneMiner Metab 30(1):1–9.

19. Liu S, et al. (2006) Pathogenic role of Fgf23 in Hyp mice. Am J Physiol EndocrinolMetab 291(1):E38–E49.

20. Consortium TH; The HYP Consortium (1995) A gene (PEX) with homologies to en-dopeptidases is mutated in patients with X-linked hypophosphatemic rickets. NatGenet 11(2):130–136.

21. Lorenz-Depiereux B, Schnabel D, Tiosano D, Häusler G, Strom TM (2010) Loss-of-function ENPP1 mutations cause both generalized arterial calcification of infancyand autosomal-recessive hypophosphatemic rickets. Am J Hum Genet 86(2):267–272.

22. Feng JQ, et al. (2006) Loss of DMP1 causes rickets and osteomalacia and identifiesa role for osteocytes in mineral metabolism. Nat Genet 38(11):1310–1315.

23. Bergwitz C, Jüppner H (2010) Regulation of phosphate homeostasis by PTH, vitamin D,and FGF23. Annu Rev Med 61:91–104.

24. White KE, et al. (2000) Autosomal dominant hypophosphataemic rickets is associatedwith mutations in FGF23. Nat Genet 26(3):345–348.

25. White KE, et al. (2001) Autosomal-dominant hypophosphatemic rickets (ADHR) mu-tations stabilize FGF-23. Kidney Int 60(6):2079–2086.

26. Benet-Pagès A, Orlik P, Strom TM, Lorenz-Depiereux B (2005) An FGF23 missensemutation causes familial tumoral calcinosis with hyperphosphatemia. HumMol Genet14(3):385–390.

27. Topaz O, et al. (2004) Mutations in GALNT3, encoding a protein involved in O-linkedglycosylation, cause familial tumoral calcinosis. Nat Genet 36(6):579–581.

28. 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.

29. Bergwitz C, et al. (2009) Defective O-glycosylation due to a novel homozygous S129Pmutation is associated with lack of fibroblast growth factor 23 secretion and tumoralcalcinosis. J Clin Endocrinol Metab 94(11):4267–4274.

30. Wang X, et al. (2010) Expression of FAM20C in the osteogenesis and odontogenesis ofmouse. J Histochem Cytochem 58(11):957–967.

31. Cong L, et al. (2013) Multiplex genome engineering using CRISPR/Cas systems. Science339(6121):819–823.

32. Jinek M, et al. (2012) A programmable dual-RNA-guided DNA endonuclease inadaptive bacterial immunity. Science 337(6096):816–821.

33. Gill DJ, Clausen H, Bard F (2011) Location, location, location: New insights intoO-GalNAc protein glycosylation. Trends Cell Biol 21(3):149–158.

34. Simpson MA, et al. (2009) Mutations in FAM20C also identified in non-lethal osteo-sclerotic bone dysplasia. Clin Genet 75(3):271–276.

35. Xiao J, Tagliabracci VS, Wen J, Kim SA, Dixon JE (2013) Crystal structure of the Golgicasein kinase. Proc Natl Acad Sci USA 110(26):10574–10579.

36. Farrow EG, et al. (2011) Iron deficiency drives an autosomal dominant hypophosphatemicrickets (ADHR) phenotype in fibroblast growth factor-23 (Fgf23) knock-in mice. ProcNatl Acad Sci USA 108(46):E1146–E1155.

37. Imel EA, et al. (2011) Iron modifies plasma FGF23 differently in autosomal dominanthypophosphatemic rickets and healthy humans. J Clin Endocrinol Metab 96(11):3541–3549.

38. Bhattacharyya N, et al. (2012) Mechanism of FGF23 processing in fibrous dysplasia.J Bone Miner Res 27(5):1132–1141.

39. Knutson MD (2010) Iron-sensing proteins that regulate hepcidin and enteric ironabsorption. Annu Rev Nutr 30:149–171.

40. Chefetz I, et al. (2009) GALNT3, a gene associated with hyperphosphatemic fa-milial tumoral calcinosis, is transcriptionally regulated by extracellular phosphateand modulates matrix metalloproteinase activity. Biochim Biophys Acta 1792(1):61–67.

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

42. Hart GW, Slawson C, Ramirez-Correa G, Lagerlof O (2011) Cross talk between O-GlcNA-cylation and phosphorylation: Roles in signaling, transcription, and chronic disease. AnnuRev Biochem 80:825–858.

43. Yang WH, et al. (2006) Modification of p53 with O-linked N-acetylglucosamine reg-ulates p53 activity and stability. Nat Cell Biol 8(10):1074–1083.

44. Tibaldi E, et al. (2010) Golgi apparatus casein kinase phosphorylates bioactive Ser-6 ofbone morphogenetic protein 15 and growth and differentiation factor 9. FEBS Lett584(4):801–805.

45. Saito S, Yano K, Sharma S, McMahon HE, Shimasaki S (2008) Characterization of thepost-translational modification of recombinant human BMP-15 mature protein. ProteinSci 17(2):362–370.

46. George A, Veis A (2008) Phosphorylated proteins and control over apatite nucleation,crystal growth, and inhibition. Chem Rev 108(11):4670–4693.

6 of 6 | www.pnas.org/cgi/doi/10.1073/pnas.1402218111 Tagliabracci et al.

Supporting InformationTagliabracci et al. 10.1073/pnas.1402218111SI MethodsMolecular Biology, Cell Culture, and Transfection.Human cDNAs forFGF23, PCSK1, PCSK3 (furin), and the family with sequence sim-ilarity 20, member C (Fam20C) were fromOpen Biosystems. HumanPCSK2 and N-acetylgalactosaminyltransferase 3 (GalNAc-T3)cDNAs were from DNASU. The ORFs were amplified by PCRand cloned into mammalian expression plasmids containinga C-terminal V5/His tag (pcDNA4.0; Invitrogen) or a C-terminalFlag tag (pCCF). QuikChange site-directed mutagenesis (Agi-lent Technologies) was performed to introduce mutations. U2OSand HEK293T cells were grown in DMEM containing 10%(vol/vol) FBS with 100 μg/mL penicillin/streptomycin (GIBCO) at37 °C with 5% CO2. For coexpression experiments, 5 × 105 cellswere seeded in 2 mL in a six-well plate format. Approximately 24 hlater, cells were transfected with 0.5 μg pCCF-PCSK1/2/3 or pCCF-Fam20C (or mutants) and 2 μg pcDNA4.0-FGF23 (or mutants)with 7 μL FuGENE 6 (Roche) as recommended by the man-ufacturer. Forty to 48 h after transfection, the conditioned me-dium and cell extracts were harvested and analyzed.

Protein Immunoblotting and Immunoprecipitations. V5- and Flag-tagged proteins were analyzed by immunoprecipitation and im-munoblotting as described (1). Furin protein was analyzed byimmunoblotting cell extracts with an anti-furin polyclonal anti-body (1:1,000 dilution; GeneTex). To detect Fam20C protein inconditioned medium, cells were extensively washed with PBSand serum-free DMEM and then incubated for a minimum of16 h in serum-free DMEM. The medium was centrifuged at 750 × gfor 5 min to remove cell debris. The supernatant was furthercentrifuged at 10,000 × g for 10 min, and 100% trichloroaceticacid (one-quarter the volume of the medium) was added to theresulting supernatant. The proteins were precipitated overnightat 4 °C and washed two or three times with −20 °C acetone. Thesamples were dried in a SpeedVac and resuspended in 1× SDSloading buffer. Fam20C protein was analyzed by immunoblottingwith a rabbit polyclonal anti-Fam20C antibody.

Generation of Anti-Fam20C Antibody. Anti-Fam20C antibodies wereaffinity-purified by coupling maltose-binding protein (MBP) fusionpeptides to HiTrap NHS-activated HP columns (GE Healthcare).The N-terminal MBP-fusion peptides (residues 20–361, 381–496,423–584, and 381–584 of human Fam20C) were produced in Es-cherichia coli and affinity-purified using amylose resin (NewEngland Biolabs).

Metabolic Radiolabeling of U2OS Cells. For metabolic radiolabelingexperiments, 5 × 105 cells were seeded in 2 mL in a six-well plateformat. Approximately 24 h later, cells were transfected with5 μg of pcDNA4.0-FGF23 or pcDNA4.0-osteopontin with 10 μLFuGENE 6. Forty to 48 h after transfection, the medium wasreplaced with phosphate-free DMEM containing 10% (vol/vol)dialyzed FBS and 1 mCi/mL [32P]orthophosphate (PerkinElmer).The cells were incubated for an additional 6–8 h at which timethe conditioned medium was centrifuged at 750 × g for 5 minto remove cell debris. The medium was further centrifuged at10,000 × g for 10 min and V5-tagged proteins were immuno-precipitated from the supernatant, washed five times with PBScontaining 0.4 mM EDTA and 1% Nonidet P-40, and thenanalyzed for protein and incorporated 32P by immunoblottingand autoradiography.

Clustered Regularly Interspaced Short Palindromic Repeats/Cas9Genome Editing. The 20-nt guide sequences targeting humanFAM20C and FURIN were designed using the clustered regularlyinterspaced short palindromic repeats (CRISPR) design toolat www.genome-engineering.org/crispr (2) and cloned into a bi-cistronic expression vector (pX330) containing human codon-optimized Cas9 and the RNA components (2) (Addgene).The guide sequences targeting exon 1 of human FAM20C and

exon 7 of human FURIN are shown below.FAM20C.

5′-GGGCTGCGCGCACGAACAGC-3′ (clone 5)

5′-CCGCCCCCGCAAGGCGCGCT-3′ (clone 9)

FURIN.

5′-TACACCACAGACACCGTTGT-3′ (clones 17 and 82)

The single-guide RNAs (sgRNAs) in the pX330 vector (4 μg)were mixed with EGFP (1 μg; Clontech) and cotransfected intoU2OS cells using FuGENE 6. Twenty-four hours posttransfection,the cells were trypsinized, washed with PBS, and resuspendedin fluorescence-activated cell sorting (FACS) buffer (PBS, 5mM EDTA, 2% FBS, and 100 μg/mL penicillin/streptomycin).GFP-positive cells were single-cell–sorted by FACS (HumanEmbryonic Stem Cell Core, University of California, San Diego;BD Influx) into a 96-well plate format into DMEM containing20% FBS and 100 μg/mL penicillin/streptomycin. Single cloneswere expanded and screened for Fam20C and furin by proteinimmunoblotting. Genomic DNA (gDNA) was purified fromclones using the Quick-gDNA Prep Kit (Zymo Research), andthe region surrounding the protospacer adjacent motif (PAM)was amplified with Q5 polymerase (New England Biolabs) usingthe following primers (linkers containing EcoRI and HindIIIrestriction sites are underlined).FAM20C.

Forward: 5′-AAAAGAATTCTGGAGAGGAGCGCGCTGA-GGATC-3′

Reverse: 5′-AAAAAAGCTTTCCGGGTTCTCCGCCGCTT-TG-3′

FURIN.

Forward: 5′-AAAAGAATTCACTCAGGGGATGATGGG-TGTC-3′

Reverse: 5′-AAAAAAGCTTAGAGAAGGAAAAAGAGA-ACACCTCC-3′

PCR products were purified using the DNA Clean & Con-centrator Kit (Zymo Research) and cloned into pBluescript IIKS+. To determine the indels of individual alleles, ∼10 bacterialcolonies were expanded and the plasmid DNA was purified andsequenced.

Protein Purification.Flag-tagged Fam20C and FGF23 R176Q wereimmunopurified from conditioned medium of HEK293T cells aspreviously described (1). FGF23 R176Q was further purified bySuperdex 200 size-exclusion chromatography. To generate a se-creted form of GalNAc-T3 (sGalNAc-T3), the nucleotides en-coding residues 38–633 of human GalNAc-T3 were cloned intoa modified retroviral (pQCXIP; Clontech) vector containingan N-terminal interleukin 2 signal peptide and a C-terminal Flag

Tagliabracci et al. www.pnas.org/cgi/content/short/1402218111 1 of 10

tag. Stable expression and purification of sGalNAc-T3 fromconditioned medium were performed as described (1).

Enzyme Assays. Kinase assays. In vitro kinase assays were performedessentially as described (1). The reaction mixture contained 32.5mM Tris·HCl (pH 7.5), 100 mM NaCl, 12.5% glycerol, 10 mMMnCl2, 0.5 mM [γ-32P]ATP (specific activity 100–500 cpm/pmol),0.5 mg/mL FGF23 R176Q, and 20 μg/mL Fam20C-Flag or Fam20C(D478A)-Flag.Glycosylation assays. In vitro O-glycosylation assays were per-formed essentially as described (3). Reactions were performed ina 20-μL mixture containing 25 mM sodium cacodylate (pH 7.4),10 mM MnCl2, 1.5 mM UDP-GalNAc (Sigma), 0.5 mg/mL pep-tide substrate: PIPRRHTRSAEDDSERDPL; FGF23(172–190)or PIPRRHTRpSAEDDSERDPL; pFGF23(172–190), and 50μg/mL sGalNAc-T3. Reactions were incubated at 37 °C andterminated at the indicated time points by the addition of anequal volume of Sigma-Aldrich universal MALDI matrix re-suspended in 78% acetonitrile and 0.1% trifluoroacetic acid.Protease assays. Furin cleavage assays were performed as previ-ously described (3). Reactions were performed in a 20-μLmixturecontaining 50 mM Hepes (pH 7.5), 1 mM CaCl2, 0.5 mg/mLFGF23(172–190) or pFGF23(172–190), and 0.02 units per μLfurin (New England Biolabs). Reactions were incubated at 37 °Cand terminated at the indicated time points by the addition ofan equal volume of Sigma-Aldrich universal MALDI matrix re-suspended in 78% acetonitrile and 0.1% trifluoroacetic acid.Deglycosylation and λ-phosphatase assays. V5-immunoprecipitatesfrom conditionedmedium of U2OS cells transiently expressing V5-tagged FGF23 were treated with O-glycosidase and α-(2→3,6,8,9)-neuraminidase to remove O-linked glycosylation or PNGase F toremove N-linked glycosylation according to the manufacturer’sinstructions (Sigma-Aldrich; Enzymatic Protein DeglycosylationKit). λ-Phosphatase assays were performed as described (1).

Matrix-Assisted Laser Desorption/Ionization Analysis. Glycosylationand protease reaction products (1 μL) were plated on an AppliedBiosciences (ABI) MALDI target plate for analysis. The ABI4800 MALDI-TOF/TOF was calibrated at 25 ppm mass errorwith a peptide mix standard. After calibration, each spot wasanalyzed using reflectron positive ionization and a laser power of30%. Masses that corresponded to truncated forms of peptidePIPRRHTRSAEDDSERDPL and the phosphorylated versionPIPRRHTRpSAEDDSERDPL were sequenced de novo forvalidation. In addition, b- and y-ion masses were used to localizephosphorylated residues on fragment ions.

Peptide Synthesis.Peptides and phosphopeptides were synthesizedby standard O-(Benzotriazol-1-yl)-N,N,N′,N′-tetramethyluroniumhexafluorophosphate/1-Hydroxybenzotriazole (HBTU/HOBt) Fmoc

solid-phase chemistry on Wang resin using an ABI 431A synthesizerand incorporating pSer as its Fmoc-O-benzyl ester. Peptide–resinwas cleaved with reagent K (2-Ethyl-5-phenylisoxazolium-3′-sulfonate). Identity and purity were assessed by electrosprayionization-Fourier transform ion cyclotron resonance mass spec-trometry; the purity of crude peptide was at least 90%. Peptideswere dissolved in 10 mM Hepes and the pH was adjusted to 7 withNaOH before use.

Mass Spectrometry. Fifty picomoles of Flag-tagged FGF23 R176Qwas oxidized with 4 mM DTT and alkylated with 8 mM iodoa-cetamide and digested with either trypsin, endoproteinase GluC,or chymotrypsin. The resulting peptide extract was diluted intoa solution of 2% acetonitrile, 0.1% formic acid (buffer A) foranalysis. Two picomoles of each digest was analyzed by automatedmicrocapillary liquid chromatography-tandem mass spectrome-try. Fused-silica capillaries (100-μm inner diameter; i.d.) werepulled using a P-2000 CO2 laser puller (Sutter Instruments) toa 5-μm i.d. tip and packed with 10 cm of 5-μm Magic C18 ma-terial (Agilent) using a pressure bomb. This column was theninstalled in-line with a Dionex 3000 HPLC pump running at 300nL/min. Peptides were loaded with an autosampler directly ontothe column and were eluted from the column by applying a30-min gradient from 5% buffer B to 40% buffer B (98% aceto-nitrile, 0.1% formic acid). The gradient was switched from 40% to80% buffer B over 5 min and held constant for 3 min. Finally, thegradient was changed from 80% buffer B to 100% buffer A over0.1 min, and then held constant at 100% buffer A for 15 moreminutes. The application of a 1.8-kV distal voltage electro-sprayed the eluting peptides directly into an LTQ XL ion trapmass spectrometer equipped with a nano-liquid chromatographyelectrospray ionization source. Full mass spectra were recorded onthe peptides over 400–2,000 m/z, followed by five tandem mass(MS/MS) events on the five most intense ions. Mass spectrometerscan functions and HPLC solvent gradients were controlled by theXcalibur data system (Thermo Finnigan). MS/MS spectra wereextracted with ReAdW.exe (http://sourceforge.net/projects/sashimi).The resulting mzXML file contains all of the data for all MS/MSspectra and can be read by the subsequent analysis software. TheMS/MS data were searched with Inspect (4) against a databasecontaining protein sequences for all E. coli proteins, commoncontaminants, and the sequence for FGF23 R176Q (2,786 pro-teins) with modifications: +16 on methionine (oxidation), +57on cysteine (carbamidomethyl), and + 80 on threonine, serine, ortyrosine (phosphorylation). Only peptides with at least a P valueof 0.01 were analyzed further. The MS/MS data of putativephosphorylated peptides were manually verified. In addition,further verification was performed by analyzing putative phos-phorylated peptides in targeted MS/MS mode.

1. Tagliabracci VS, et al. (2012) Secreted kinase phosphorylates extracellular proteins thatregulate biomineralization. Science 336(6085):1150–1153.

2. Hsu PD, et al. (2013) DNA targeting specificity of RNA-guided Cas9 nucleases. NatBiotechnol 31(9):827–832.

3. Kato K, et al. (2006) Polypeptide GalNAc-transferase T3 and familial tumoral calcinosis.Secretion of fibroblast growth factor 23 requires O-glycosylation. J Biol Chem 281(27):18370–18377.

4. Tanner S, et al. (2005) InsPecT: Identification of posttranslationally modified peptidesfrom tandem mass spectra. Anal Chem 77(14):4626–4639.

Tagliabracci et al. www.pnas.org/cgi/content/short/1402218111 2 of 10

300 400 500 600 700 800 900 1000 1100 1200 1300 1400m/z

0

10

20

30

40

50

60

70

80

90

100632.0926

775.4199660.3980 716.4478831.4466

488.3409 574.5496 991.4409946.4575 1233.53811165.5139416.3833309.2543 862.4384 1075.5848 1379.69821314.6639

300 400 500 600 700 800 900 1000 1100 1200 1300 1400m/z

0

10

20

30

40

50

60

70

80

90

100737.1538

623.0914672.0321

1342.4809831.4779776.9341 946.5019500.4369 1146.54851076.9574565.5755 1244.5568398.3367327.3252 1324.4948984.6955894.5484 1384.7170

y10+2

y6 y7 y8 y9 y11

178T R S A E D D S E R D P L190

y11 y10 y9 y8 y7 y6

b6 b7 b8 b11b9

b6

b11+2

178T R Sp A E D D S E R D P L190

y8 y7 y4

b7 b11 b12

b7

b8b9

b11+2

(b11-H3PO4)+2

(MH-H3PO4)+2

y4y7 y8 (b11-H3PO4)+2

(b12-H3PO4)+2

b7

A

B

Fig. S1. FGF23 is phosphorylated on Ser180. Representative MS/MS fragmentation spectra of a chymotryptic peptide (FGF23 178–190) depicting Ser180

phosphorylation of FGF23 R176Q purified from conditioned medium of HEK293T cells. The nonphosphopeptide and the phosphopeptide are shown in A and B,respectively.

RT: 42.00 - 57.00 SM: 15G

42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57Time (min)

0

50

100

0

50

100

0

50

100RT: 45.68AA: 812455

RT: 46.99AA: 38501

RT: 51.81AA: 2432500

RT: 42.00 - 57.00 SM: 15G

42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57Time (min)

0

50

100

0

50

100

0

50

100 RT: 46.16AA: 714375

RT: 47.47MA: 340573

48.55 49.36 50.24 50.91 51.49 52.18 52.65 53.2045.9445.58 55.4046.49 55.7354.88 56.8143.8843.1642.70

RT: 52.45AA: 600174

SAEDDSERDPLNVLKPR

SphosAEDDSERDPLNVLKPR

SEDAGFVVITGVMSR for loading control

SAEDDSERDPLNVLKPR

SphosAEDDSERDPLNVLKPR

SEDAGFVVITGVMSR for loading control

- Fam

20C

+F

am20

C

A

B

Fig. S2. Fam20C phosphorylates FGF23 on Ser180. Selected ion chromatograms of tryptic peptides (FGF23 180–196) from FGF23 R176Q (A) or FGF23 R176Q thathad been treated with recombinant Fam20C (B). Note the relative 10-fold increase in the relative abundance of the phosphopeptide upon treatment withFam20C. Calculations are estimated based on the areas under the curve for the selected ion species (AA phospho/nonphospho; 38,501/812,455 = 0.05; 340,573/714,375 = 0.5). AA and MA, area; RT, retention time.

Tagliabracci et al. www.pnas.org/cgi/content/short/1402218111 3 of 10

5’...GCCGAGCCGGCCGAGCGCGCCTTGCGGGGGCGGGATCCCGGCGCCC...3’

5’...GCCGATCCT------------TGCGGGGGCGGGATCCCGGCGCCC...3’

5’...GCCGAGCCGCCCCGCAA---------------------GGCGCCC...3’

3’...CGGCTCGGCCGGCTCGCGCGGAACGCCCCCGCCCTAGGGCCGCGGG...5’PAM

3’-UCGCGCGGAACGCCCCCGCC-5’ sgRNA_05

Allele 1

Allele 2

Human locus

GCCGAGCCGCCCCGCAAGGCGCCC

GCCGATCCTTGCGGGGGCGGGATCCCGGCGCCC

1 22 584

SP Kinase Domain

Frameshift116

122Stop

1 22584

SP Kinase Domain

Frameshift116

125Stop

Indel

A

Indel

5’

3’ 5’

3’

sgRNA

Cas9

3’

5’

PAM

G

G N

20 nt

B

5’...TCGGGGGAGCCCGGCG-----------------------GAGGTGG...3’

5’...TCGGGGGAGCCCGGCTGTTCGTGCGCGCAGCCCGCCGCCGAGGTGG...3’

3’...AGCCCCCTCGGGCCGACAAGCACGCGCGTCGGGCGGCGGCTCCACC...5’

Human locus

PAM

3’-CGACAAGCACGCGCGUCGGG-5’ sgRNA_02

TCGGGGGAGCCCGGCGGAGGTGG 1 22 584

SP Kinase Domain

Frameshift46

200Stop

IndelC

D

E

Fig. S3. Generation of Fam20C knockout cells using CRISPR/Cas9 genome editing. (A) The type II prokaryotic CRISPR/Cas9 from Streptococcus pyogenes hasbeen shown to facilitate RNA-guided site-specific DNA cleavage and can be harnessed to target the destruction of specific genes in mammalian cells (1–3). Anygenomic locus followed by a 5′-NGG PAM (red) can be targeted by a chimeric single guide RNA (sgRNA) (green) consisting of a 20-nt guide sequence anda scaffold. The sgRNA directs the Cas9 nuclease (gray) to the genomic target, resulting in a double-strand break. The double-strand break is repaired by error-prone nonhomologous end joining or by homologous recombination. (B and D) Schematic representations of the base pairing between guide RNAs and thetargeting locus of exon 1 in the human FAM20C gene. (C and E) The sequences of the mutated alleles in FAM20C clone 5 (C) and clone 9 (E) and representativechromatograms depicting the indels (red) are shown. The indels are predicted to cause frameshift mutations producing inactive copies of the protein. SP, signalpeptide.

1. Cong L, et al. (2013) Multiplex genome engineering using CRISPR/Cas systems. Science 339(6121):819–823.2. Jinek M, et al. (2012) A programmable dual-RNA-guided DNA endonuclease in adaptive bacterial immunity. Science 337(6096):816–821.3. Jinek M, et al. (2013) RNA-programmed genome editing in human cells. Elife 2:e00471.

Tagliabracci et al. www.pnas.org/cgi/content/short/1402218111 4 of 10

100 KDa

55 KDa

70 KDa

35 KDa

130 KDa

250 KDa

sGalNAc-T3

Fig. S4. Expression and purification of sGalNAc-T3. SDS/PAGE and Coomassie staining of Flag-tagged sGalNAc-T3 immunopurified from the conditionedmedium of HEK293T cells.

Tagliabracci et al. www.pnas.org/cgi/content/short/1402218111 5 of 10

800 1180 1560 1940 2320 2700M ass (m/z )

1298 .1

0102030405060708090

100

% I

nte

ns

i

2246 .9456 2450 .0085

2145 .9033

875 .9642 2432 .00422090 .8511 2228 .94951989 .8335 2573 .99712408 .01951078 .0222 2161 .8525 2302 .98051852 .80471225 .5336 1954 .8719889 .0198 1358 .4662 1696 .61661474 .6487

800 1180 1560 1940 2320 2700M ass (m/z )

3481 .8

0

20

40

60

80

100

% I

nte

ns

ity

ty

2246 .8987

2145 .88212449 .9575

2090 .8315 2228 .94091989 .7943 2462 .9995875 .9891 1124 .4659 2361 .94682127 .89921852 .7705 2239 .75421225 .9937 1954 .7197995 .2341 1325 .2849 1696 .65651448 .3438 1596 .4999

800 1180 1560 1940 2320 2700M ass (m/z )

589 .6

0

20

40

60

80

100

% I

nte

ns

ity

2246 .9341

875 .95642145 .8850 2268 .8789867 .0152

2450 .03471078 .0046 2272 .06031989 .7593 2134 .86451289 .0570888 .9924 1646 .4606 1785 .78361438 .4230801 .0538 2538 .93071538 .6671

800 1180 1560 1940 2320 2700M ass (m/z )

2 .2E+4

0

20

40

60

80

100

% I

nte

ns

ity

2246 .9653

2145 .93432268 .94261989 .8309 2127 .98141852 .7721875 .9688 1416 .68141124 .4824 2464 .10571531 .6946 1696 .65921301 .6438995 .8828

PIPRRHTRSAEDDSERDPL

PIPRRHTRSAEDDSERDPL

PIPRRHTRSAEDDSERDPL

PIPRRHTRSAEDDSERDPL

PIPRRHTRSAEDDSERDPL +

203 Da (galNAc)

PIPRRHTRSAEDDSERDPL +

203 Da (galNAc)

PIPRRHTRSAEDDSERDPL +

203 Da (galNAc)

800 1180 1560 1940 2320 2700M ass (m/z )

4063 .3

0102030405060708090

100

% I

nte

ns

ity

2326 .8223

2229 .0076812 .1735 2348 .8284 2592 .89581164 .3790 2212 .0173918 .1966 2449 .90841411 .36991025 .1989 1268 .3641 1549 .3765 1806 .5597 1961 .61111657 .3813

800 1180 1560 1940 2320 2700M ass (m/z )

586 .3

0

20

40

60

80

100

% I

nte

ns

ity

2326 .8025

2329 .65432229 .76101163 .9380 2348 .7607866 .9969 2477 .7830 2592 .88382212 .68121289 .0267 1805 .59551567 .55621155 .5500 1419 .4576 1675 .3680955 .1647

800 1180 1560 1940 2320 2700M ass (m/z )

2648 .0

0

20

40

60

80

100

% I

nte

ns

ity

2326 .7622

2228 .95242348 .73391164 .4160 2593 .6543875 .9684 2210 .8323 2483 .79691411 .3496993 .8598 1298 .2729 1594 .2863 1805 .5557

800 1180 1560 1940 2320 2700M ass (m/z )

1 .9E+4

0

20

40

60

80

100

% I

nte

ns

ity

2326 .8672

2229 .89702308 .9275875 .9765 2423 .89502185 .8677 2592 .92902056 .87571164 .4697 1496 .6134995 .9240 1919 .85571642 .7976 1805 .62651357 .6862

pFGF23(172-190)t = 0 hr

PIPRRHTRSAEDDSERDPL + 80 Da

PIPRRHTRSAEDDSERDPL + 80 Da

PIPRRHTRSAEDDSERDPL + 80 Da

PIPRRHTRSAEDDSERDPL + 80 Da

A

B

pFGF23(172-190)t = 1 hr

pFGF23(172-190)t = 4 hr

pFGF23(172-190)t = o/n

FGF23(172-190)t = 0 hr

FGF23(172-190)t = 1 hr

FGF23(172-190)t = 4 hr

FGF23(172-190)t = o/n

Fig. S5. Ser180-phosphorylated FGF23(172–190) is not a substrate for sGalNAc-T3. Full MALDI-TOF MS spectra depicting the time-dependent incorporation ofGalNAc from UDP-GalNAc into FGF23(172–190) (A) or Ser180-phosphorylated FGF23(172–190) (B).

Tagliabracci et al. www.pnas.org/cgi/content/short/1402218111 6 of 10

168 612 1056 1500 1944 2388Mass (m/ z)

2058. 9

0

10

20

30

40

50

60

70

80

90

100

% I

ntens

ity

1619. 8210

1621. 7793

2221. 8337

1602. 8506 1734. 84971521. 8096

1636. 85942210. 9585

2203. 99071217. 7279 1606. 0093217. 2169 1950. 8562 2249. 1919

2123. 88061813. 69671174. 6664 1643. 03801487. 7770840. 4355 1286. 61211158. 5181578. 5170 947. 4656211. 1945 464. 4430355. 6335

P I P R R H T R S A E D D S E R D P L

+203

b2 b4

b8 b10

b11

b12

b13

b15 b16

b17

b5 y7

Fig. S6. Thr178 is O-glycosylated by sGalNAc-T3. Representative MS/MS fragmentation spectrum of the selected ion peak at m/z 2450 in Fig. S5A, depicting theaddition of GalNAc to Thr178.

Tagliabracci et al. www.pnas.org/cgi/content/short/1402218111 7 of 10

5’...GCTGCGGTGGCCAACA-------------------TGTGGCCTACA...3’

5’...GCTGCGGTGGCCAACAACGGTGTCTGTGGTGTAGGTGTGGCCTACA...3’

3’...CGACGCCACCGGTTGTTGCCACAGACACCACATCCACACCGGATGT...5’

Human locus

PAM

sgRNA_18

794

SP FURIN

Frameshift208

253Stop

5’...GCTGCGGTGGCCAAAACACGGTGTCTGTGGTGTAGGTGTGGCCTACA...3’

Allele 1

Allele 2

GCTGCGGTGGCCAACATGTGGCCTACA

Indel

3’-UGUUGCCACAGACACCACAU-5’

GCTGCGGTGGCCAAAACACGGTGTCTGTGGTGTAGGTGTGGCCTACA794

SP FURIN

Frameshift207

259Stop

Allele 1

Allele 2

Indel

A

B

250 KDa

130 KDa

100 KDa

70 KDa

55 KDa

35 KDa

-Furin

-GAPDH

Clo

ne 1

7

Con

trol

C

5’...CGGTGT-----------------------------------CTGTGGTGTAGGTGTGGC...3’

5’...CGGTGTGCGGGGGAAGTGGCTGCGGTGGCCAACAACGGTGTCTGTGGTGTAGGTGTGGC...3’

3’...GCCACACGCCCCCTTCACCGACGCCACCGGTTGTTGCCACAGACACCACATCCACACCG...5’

Human locus

PAM

sgRNA_18

5’...CGGTGTGCGGGGGAAGTGGCTGCGGTG-------------TCTGTGGTGTAGGTGTGGC...3’

Allele 1

Allele 2

Indel

3’-UGUUGCCACAGACACCACAU-5’

Indel

D

CGGTGTCTGTGGTGTAGGTGTGGC

CGGTGTGCGGGGGAAGTGGCTGCGGTGTCTGTGGTGTAGGTGTGGC

Allele 1

Allele 2

794

SP FURIN

Frameshift199

247Stop

794

SP FURIN

Frameshift206

209Stop

Clone 17

Clone 82

Clo

ne 8

2

Extracts

E

Fig. S7. Generation of FURIN knockout cells using CRISPR/Cas9 genome editing. (A and C) Schematic representations of the base pairing between a guide RNA(sgRNA_18) and the targeting locus of exon 7 in the human FURIN gene. (B and D) The sequences of the mutated alleles in FURIN clone 17 (B) and clone 82 (D)and representative chromatograms depicting the indels (red) are shown. The indels are predicted to cause frameshift mutations producing inactive copies ofthe protein. (E) Protein immunoblotting of cell extracts from control and furin KO cells.

Tagliabracci et al. www.pnas.org/cgi/content/short/1402218111 8 of 10

800 1140 1480 1820 2160 2500M ass (m/z )

1 .4E+4

0102030405060708090

100

% I

nte

ns

i ty

1032 .562

931 .5212090 .8981054 .550 1989 .861 2246 .9771233 .455 1852 .813876 .464 953 .513 1138 .578803 .195 1696 .7181472 .728 2385 .0521323 .307

800 1140 1480 1820 2160 2500M ass (m/z )

6 .2E+4

0

20

40

60

80

100

% I

nte

ns

ity

1032 .539

931 .5132090 .898

1054 .545 1989 .8612246 .9801015 .509 2161 .9452072 .9211233 .457 1852 .8041122 .601876 .468 1971 .880803 .188 947 .502 2326 .9411472 .727 1696 .6821315 .644 1583 .629

800 1140 1480 1820 2160 2500M ass (m/z )

5 .6E+4

0102030405060708090

100

% I

nte

ns

ity

1032 .555

2246 .9922145 .955

2229 .0221989 .8611054 .541 2127 .989931 .513 1852 .800803 .180 2314 .0211233 .447 1416 .693 1696 .7071318 .6951138 .583 1531 .705

800 1140 1480 1820 2160 2500M ass (m/z )

7076 .1

0102030405060708090

100

% I

nte

ns

ity

2246 .990

2145 .951 2229 .0171124 .505 2206 .1021989 .856 2314 .0382109 .8951852 .791812 .207 918 .240 1416 .715 2396 .0121283 .3271015 .257 1726 .5371203 .345 1592 .3431505 .381

PIPRRHTRSAEDDSERDPL

PIPRRHTRSAEDDSERDPL

PIPRRHTRSAEDDSERDPL

PIPRRHTRSAEDDSERDPL

PIPRRHTR

PIPRRHTR

PIPRRHTR

SAEDDSERDPL

SAEDDSERDPL

SAEDDSERDPL

-h20-101-156

-257

-101

-101

-101

PIPRRHT

PIPRRHT

800 1180 1560 1940 2320 2700M ass (m/z )

1 .1E+4

0102030405060708090

100

% I

nte

ns

ity

1032 .7383

931 .5189

1188 .6627933 .5178 1054 .5577 1169 .6224 1651 .9218850 .4355 1298 .6570 1788 .9728990 .5574 1495 .83341184 .5713890 .4687 1301 .6627 2314 .0679 2464 .0020 2593 .04982157 .9851800 .1817 1961 .79661609 .9100 1808 .0243

800 1180 1560 1940 2320 2700M ass (m/z )

6 .8E+4

0

20

40

60

80

100

% I

nte

ns

ity

1032 .5347

1188 .64621054 .5345

931 .5107 1145 .6257 1651 .90121298 .6360 1495 .8112 2225 .8679 2326 .9136803 .1787 1030 .5420 1788 .9540 2463 .9319 2592 .94822056 .86301919 .8431

800 1180 1560 1940 2320 2700M ass (m/z )

3 .0E+4

0

20

40

60

80

100

% I

nte

ns

ity

1032 .5424

1054 .5382 1188 .6511 2326 .90581651 .8958803 .1926 931 .5209 1076 .5306 1788 .9536 2225 .88921495 .81071184 .5566 1298 .6267 2592 .98612463 .97852334 .16332056 .88381919 .7991

800 1180 1560 1940 2320 2700M ass (m/z )

4477 .7

0

20

40

60

80

100

% I

nte

ns

ity

2326 .8293

2230 .35572227 .94021164 .4210813 .2053 2592 .90582424 .8955919 .2242 1033 .2991 2056 .84941297 .4475 1411 .3760 1513 .7103 2163 .22631805 .56191658 .4591

PIPRRHTRSAEDDSERDPL + 80 Da (phosphorylation)

PIPRRHTR

PIPRRHTR

PIPRRHTR

-101

-101

-101

PIPRRHTRSAEDDSERDPL + 80 Da

t = 0 hr

pFGF23(172-190)t = 1 hr

pFGF23(172-190)t = 4 hr

pFGF23(172-190)t = o/n

FGF23(172-190)t = 0 hr

FGF23(172-190)t = 1 hr

FGF23(172-190)t = 4 hr

FGF23(172-190)t = o/n

pFGF23(172-190)

A

B

Fig. S8. Ser180-phosphorylated FGF23(172–190) is cleaved by furin. Full MALDI-TOF MS spectra depicting the time-dependent cleavage of FGF23(172–190) (A)or Ser180-phosphorylated FGF23(172–190) (B).

Tagliabracci et al. www.pnas.org/cgi/content/short/1402218111 9 of 10

Fam20C

Furin

N-TerFragment

Inactive FGF23

FGF23 FGF23

PPP

NFGF23

CFGF23

PPP

P ?

P

C

N

ATP

?

GalNAcGalNAc

Active FGF23

i)

Furin

FGF23 FGF23

C

N

ii) GalNAc-T3others?

UDP-GalNAc

GalNAc

GalNAcGalNAc

FGF23 FGF23

C

NGalNAc

X

Fig. S9. Model for the regulation of FGF23 by Fam20C. We propose that interplay between phosphorylation by Fam20C (i) and O-glycosylation by GalNAc-T3(ii) regulates the processing and activity of FGF23 by furin. Phosphorylation of FGF23 at Ser180 by Fam20C inhibits GalNAc-T3-mediated O-glycosylation at Thr178

and promotes inactivation by furin. Fam20C may also phosphorylate predicted Ser-x-Glu/pSer residues in the C-terminal region of FGF23.

Tagliabracci et al. www.pnas.org/cgi/content/short/1402218111 10 of 10