9-O-Acetylation of sialic acids is catalysed by CASD1 via a covalent acetyl-enzyme intermediate

Sialic acids, terminal sugars of glycoproteins and glycolipids, play important roles in development, cellular recognition processes and host–pathogen interactions. A common modification of sialic acids is 9-O-acetylation, which has been implicated in sialoglycan recognition, ganglioside biology, and the survival and drug resistance of acute lymphoblastic leukaemia cells. Despite many functional implications, the molecular basis of 9-O-acetylation has remained elusive thus far. Following cellular approaches, including selective gene knockout by CRISPR/Cas genome editing, we here show that CASD1—a previously identified human candidate gene—is essential for sialic acid 9-O-acetylation. In vitro assays with the purified N-terminal luminal domain of CASD1 demonstrate transfer of acetyl groups from acetyl-coenzyme A to CMP-activated sialic acid and formation of a covalent acetyl-enzyme intermediate. Our study provides direct evidence that CASD1 is a sialate O-acetyltransferase and serves as key enzyme in the biosynthesis of 9-O-acetylated sialoglycans. Supplementary information The online version of this article (doi:10.1038/ncomms8673) contains supplementary material, which is available to authorized users.

S ialic acids (Sias) are acidic nine-carbon sugars that commonly cap the glycan chains of cell surface glycoproteins and glycolipids. Sialoglycotopes are key determinants for numerous cell recognition events and play fundamental roles in development and host-pathogen interactions 1 . Of note, Sias show a pronounced structural diversity that arises from Nand O-substitutions of the parent compound neuraminic acid and this diversity is of critical importance for Sia recognition and subsequent cellular processes. For example, the prevalent human Sia 5-N-acetylneuraminic acid (Neu5Ac) can be substituted by 9-O-acetylation, a modification that affects the binding of several Sia-specific lectins and has functional implications in particular in the immune system 2-6 . 9-O-Acetylation impairs host cell binding of influenza A and B viruses, but at the same time creates cellular receptors for influenza C and several nidoviruses [7][8][9] . Another prominent example for the impact of 9-O-acetylation is given by CD22, a B-cell-restricted member of the Sia-binding immunoglobulin-type lectin (siglec) family. On binding to a2,6sialylated N-glycans, CD22 regulates B-cell antigen receptor signalling by setting a threshold for receptor activation 10 . In vitro data showed that the CD22-ligand interaction is blocked by 9-Oacetylation 11 , a process that is in vivo tightly controlled by two counteracting enzymatic activities: sialate O-acetyltransferase (SOAT) and sialate 9-O-acetylesterase (SIAE). Ablation of SIAE function by targeted gene knockout in mice resulted in increased Sia 9-O-acetylation and defective CD22 signalling in B-cells and caused defects in peripheral B-cell development and tolerance 12 . Moreover, catalytically defective rare variants of SIAE were associated with autoimmune diseases in humans 13,14 .
Sia O-acetylation is also a developmentally regulated modification of gangliosides, implicated in neural precursor cell migration, peripheral nerve regeneration and Mycobacterium leprae infection of Schwann cells [15][16][17] . Best studied is the 9-Oacetylation of GD3, a disialo ganglioside that acts as a potent inducer of apoptosis on translocation to mitochondrial membranes 18 . 9-O-Acetylation of the terminal a2,8-linked Sia residue blocks the pro-apoptotic activity of GD3 and promotes survival of cancer cells [19][20][21] . In acute lymphoblastic leukaemia (ALL), survival and drug resistance of lymphoblasts critically depend on 9-O-acetylation, which was found on both GD3 and sialoglycoproteins 22,23 . Enzymatic removal of 9-O-acetyl groups from internal and cell surface-bound Sias was lethal to ALL cells, opening new perspectives for therapeutic concepts 23 . The use of monoclonal antibodies recognizing the carbohydrate epitopes of non-O-acetylated GD3 (CD60a), 7-O-acetylated GD3 (CD60c) or 9-O-acetylated GD3 (CD60b) revealed a stimulatory or co-stimulatory effect of anti-CD60c and anti-CD60b antibodies on the proliferation of human lymphocytes and implicated distinct roles of 7-O-and 9-O-acetylated GD3 during activation and apoptosis of tonsillar B and T lymphocytes 24,25 .
Biochemical studies demonstrated that O-acetylation of Sia is a postsynthetic modification that takes place in the Golgi apparatus, probably in concert with Golgi-resident sialyltransferases and coupled with the import of acetyl groups from cytosolic acetyl-coenzyme A (acetyl-CoA), the presumptive acetyl donor 26,27 . However, the genes encoding eukaryotic SOATs have remained unknown. Despite many attempts, expression cloning was not successful [28][29][30] and isolation of SOAT activity proved difficult, mainly due to an inherent sensitivity to membrane solubilization [31][32][33][34] . Searching for human candidate genes, Arming et al. 35 identified CASD1 (CAS1 domain containing 1), which shares sequence similarity to CAS1 (Capsule synthesis 1) of the pathogenic fungus Cryptococcus neoformans. Deletion of the fungal gene revealed an essential role in mannose O-acetylation of the unique fungal capsular polysaccharide glucuronoxylomannan 36 . By contrast, little is known on the function of human CASD1. CASD1 overexpression as well as short interfering RNA knockdown assays performed in the initial study by Arming et al. indicated a role in 7-O-acetylation of GD3 (ref. 35). However, studies addressing SOAT function of a given protein by transfection studies proved, so far, difficult due to endogenous SOAT activity in many common mammalian cell lines 29,35,37 , a problem that might have impeded also previous expression cloning approaches.
On the basis of a naturally occurring 9-O-Ac-negative cell line and CASD1 knockout cells generated by CRISPR/Cas-mediated genome editing, we now show that CASD1 is essential for 9-Oacetylation of cellular sialoglycoconjugates. Insight into the catalytic mechanism of CASD1 is given by biochemical analysis of a soluble CASD1 variant encompassing the N-terminal catalytic domain. This reveals the formation of a covalent acetyl-enzyme intermediate, which involves S94 placed in a catalytic triad. Finally, we provide direct in vitro evidence for SOAT activity by demonstrating CASD1-mediated transfer of acetyl groups from acetyl-CoA to CMP-activated Neu5Ac.

Results
Homology modelling and topology of CASD1. Human CASD1 encodes a multimembrane spanning protein of 797 amino acids. For residues 83-290, homology modelling predicted a GDSL/SGNH-like a/b-fold that forms the scaffold for a catalytic triad composed of S94, D270 and H273, with S94 as part of a conserved GDS sequence motif (Fig. 1a). The acronyms of the family name GDSL/SGNH reflect the presence of four invariant residues, namely S in a GDS(L) motif, G located after strand S2, N placed in a GxND motif after strand S3 and H found in a DxxH motif N-terminal to the last helix 38 . However, the N-terminal domain of CASD1 differs from the canonical GDSL/SGNH fold by lacking the conserved residues G and N, and is therefore grouped into pfam family PF13839 (GDSL/SGNH-like acylesterase family found in Pmr5 and Cas1p) 39 .
To analyse the subcellular localization and topology of CASD1, a full-length variant, containing an N-terminal V5 and a C-terminal Myc epitope, was expressed in the murine fibroblast cell line LM-TK À . The orientation of the epitope tags was determined by selective membrane permeabilization and immunofluorescence analysis (Fig. 1b,c). Under conditions that allowed antibody access to all cellular compartments (0.2% Triton X-100), both tags were detectable and showed co-localization with the Golgi marker a-mannosidase II (a-Man II). Selective permeabilization of the plasma membrane with low concentrations of digitonin (5 mg ml À 1 ) allowed visualization of the N-terminal V5-tag, but not the C-terminal Myc-tag, demonstrating cytosolic and luminal orientation of N-and C-terminus, respectively. The integrity of the Golgi membrane was demonstrated by parallel staining with an antibody specific for the luminal domain of a-Man II 40 . Staining was observed exclusively in Triton X-100, but not in digitonin-permeabilized cells. Identical results were obtained on staining with an antibody against the SGNH-like domain of CASD1 (pAb 500), demonstrating that this domain faces the Golgi lumen (Fig. 1c). On the basis of these results and on our predictions for the CASD1 transmembrane domains ( Supplementary Fig. 1), we propose a topology model encompassing a luminal catalytic domain flanked by one N-terminal and 12 C-terminal transmembrane domains (Fig. 1a).
CASD1 forms a covalent acetyl-enzyme intermediate. To analyse the enzymatic activity of CASD1 in vitro, we generated a construct that allowed the expression of a soluble secreted form of CASD1 (termed sCASD1), which encompassed the SGNH-like luminal domain and a C-terminal Myc-His 6 -tag (Fig. 2a-c).
This form was produced in baculovirus-infected Sf9 insect cells, either as wild-type variant (sCASD1-wt) or as active site serine mutant (sCASD1-S94A). Affinity purified proteins were then used as an enzyme source for in vitro assays. First, we analysed whether sCASD1 possesses esterase activity, since the GDSL/SGNH fold is usually found among hydrolytic enzymes 38 , including the sialate 9-O-acetylesterase of Escherichia coli O157:H7 and the sialate 9-O-acetylesterase domains of the haemagglutinin-esterase (HE) proteins of influenza C and bovine coronavirus (BCoV) [41][42][43] . However, under conditions that displayed high O-acetyltransferase activity for BCoV-HE, no activity was observed for sCASD1-wt, neither towards the synthetic acetylesterase substrate para-nitrophenylacetate (pNP-acetate; Fig. 2d) nor towards natural O-acetylated sialoglycoconjugates (not shown). Thus, while the N-terminal luminal domain of CASD1 displays the characteristic fold of an esterase 39 , it does not appear to function as such. We therefore tested whether sCASD1 might catalyse the reverse reaction and act as a SOAT. Indeed, in the presence of [ 3 H]acetyl-CoA, sCASD1-wt showed acetyltransferase activity, resulting in the formation of radiolabelled protein, even in the absence of acceptor substrates (Fig. 2e). No radioactive incorporation was seen in parallel experiments with the mutant sCASD1-S94A, demonstrating that the observed acetyl transfer was enzyme catalysed. SOAT activity of sCASD1 towards its own N-glycans was improbable. Although sCASD1 is N-glycosylated (Fig. 2c), glycoproteins that are produced in Sf9 cells typically bear pauci mannosidic N-glycans lacking penultimate galactose and terminal Sia residues 44 . Consequently, we asked whether sCASD1 can catalyse the formation of a covalent acetyl-enzyme intermediate that could be trapped in the absence of acceptor substrate. Therefore, purified sCASD1 was incubated in the presence or absence of acetyl-CoA, digested with trypsin and analysed by liquid chromatography-electrospray ionization-mass spectrometry (LC-ESI-MS). After incubation of sCASD1-wt with acetyl-CoA, the peptide containing the active site residue S94 (87-HIAFIGDSR-95) occurred as a mixed population of unmodified and modified peptide with a mass shift of 42 Da (shift by m/z ¼ 21 for the doubly charged ion), consistent with the formation of an acetyl adduct ( Fig. 2f; note that the sequence of the peptides was confirmed by tandem MS/MS; see Supplementary Fig. 2). Analysis of the peak areas in the extracted ion chromatograms revealed that about 19% of the peptide population was acetylated ( Supplementary Fig. 3). In parallel experiments with sCASD1-S94A, only the respective unmodified peptide (87-HIAFIGDAR-95) was found ( Fig. 2f;  To test SOAT activity, purified sCASD1 was incubated with acetyl-CoA and Sia-containing acceptor substrates. After the reaction, Sia was released by acidic hydrolysis to allow derivatization with 1,2-diamino-4,5-methylenedioxy-benzene (DMB) at the free reducing end. The obtained fluorescent Sia derivatives were analysed by reversed-phase high-performance liquid chromatography (HPLC; Fig. 3a-e). SOAT activity was clearly detected towards CMP-Neu5Ac (as indicated by an additional peak in the HPLC profile, which eluted at the same retention time as DMBderivatized 9-O-acetylated Neu5Ac (Neu5,9Ac 2 ) of the reference panel; see Fig. 3a,c). The additional peak was not detected in samples obtained from parallel reactions without enzyme (Fig. 3b) or from reactions with sCASD1-S94A (Fig. 3d).
Compared with free Neu5Ac or sialoglycoconjugates, CMP-Neu5Ac was by far the best acceptor substrate (Fig. 3e). The identity of the product peak obtained with CMP-Neu5Ac was confirmed by LC-ESI-MS/MS, which showed the expected parent ion mass of DMB-labelled mono-O-acetylated Neu5Ac and a fragmentation pattern characteristic for DMB-Neu5,9Ac 2 (ref. 45; Fig. 3f,g). Together, these data demonstrate that the isolated SGNH-like domain of CASD1 displays SOAT activity in vitro and transfers acetyl groups from acetyl-CoA to CMP-Neu5Ac, resulting in the formation of CMP-Neu5,9Ac 2 .
CASD1 mediates 9-O-acetylation of cellular sialoglycans. To study the enzymatic function of CASD1 in a cellular context, we first screened for a mammalian cell line that lacks O-acetylated sialoglycans. As a detection tool, we made use of the haemagglutinin-esterase (HE) protein of BCoV. BCoV uses 9-Oacetylated sialoglycans as host cell receptors and the HE contains   Importantly, all cell lines that were positive for 9-O-acetylation were also positive for CASD1 transcripts ( Supplementary Fig. 4b).
In fact, we found only a single cell line, the murine fibroblast cell line LM-TK À , that was devoid of both ( Supplementary Fig. 4). Transfection of this naturally occurring O-Ac-negative cell line with CASD1 complementary DNA (cDNA), but not empty vector, led to the formation of Golgi-resident 9-O-acetylated   In all profiles, the retention time corresponding to DMB-Neu5,9Ac 2 of the reference panel is highlighted by a grey box. (e) SOAT activity of sCASD1 towards different acceptor substrates. Purified sCASD1wt and sCASD1-S94A were incubated with acetyl-CoA and the indicated acceptor substrates. Product analysis was performed by DMB-HPLC analysis and the amount of DMB-Neu5,9Ac 2 was quantified by integrating the corresponding peak areas. Parallel HPLC runs of reactions without CASD1 were used to measure the background, and data are presented as background subtracted data with the values obtained for CMP-Neu5Ac in the presence of sCASD1-wt and acetyl-CoA set to 100% (mean ± s.d., n  (Fig. 4a). De-Oacetylation, by either alkali treatment or the sialate 9-Oacetylesterase activity of BCoV-HE, completely abolished the virolectin signals and was used as specificity control (Fig. 4a). After stable expression of wild-type CASD1, 9-O-acetylated sialoglycotopes were detected in all cells and co-localized with the Golgi marker a-Man II and Golgi-localized V5-tagged CASD1 (Fig. 4b). Introduction of the S94A exchange into full-length CASD1 (CASD1-S94A) resulted in a loss of function. Although the mutant protein was expressed and localized to the Golgi, virolectin staining was seen only in o1% of the cells (Fig. 4c). Taken together, these experiments demonstrate that CASD1 mediates sialate 9-O-acetylation in LM-TK À cells and that this function is impaired by alanine exchange of S94.
CASD1 knockout abolishes Sia O-acetylation in HAP1 cells. To confirm that CASD1 plays a crucial role in the formation of 9-Oacetylated sialoglycans, we generated a CASD1 knockout in the near-haploid human cell line HAP1 using the CRISPR/Cas9 (clustered regularly interspaced short palindromic repeats/ CRISPR-associated 9) system 48 . The obtained knockout cell line (HAP1 DCASD1 ) harbours a 16-bp frame shift mutation in exon 3 of CASD1, allowing only the production of a truncated polypeptide that already lacks S94, the most N-terminally located residue of the catalytic triad ( Fig. 5a-c). Whereas parental HAP1 cells expressed 9-O-acetylated sialoglycans in the Golgi apparatus, 9-O-acetylation was completely lost in HAP1 DCASD1 cells (Fig. 5d). This was further confirmed by DMB-HPLC analysis revealing the presence of DMB-derivatized Neu5,9Ac 2 in parental but not in HAP1 DCASD1 cells ( Supplementary Fig. 6).
Transfection with CASD1 cDNA, but not empty vector, resulted in successful complementation of the loss-of-function defect and restored 9-O-acetylation of Golgi-localized sialoglycotopes (Fig. 5e). Of note, a genetic knockout carried out in parallel in human embryonic kidney (HEK) 293T cells showed the same outcome ( Supplementary Fig. 7) and thus provided independent evidence for the essential role of CASD1 in Sia 9-O-acetylation.
CASD1 mediates 9-O-acetylation of gangliosides. On the basis of HAP1 and HAP1 DCASD1 cells, we next explored the role of CASD1 in the 9-O-acetylation of gangliosides such as GD3. To do so, we made use of a set of monoclonal antibodies (mAbs) that distinguish between GD3 (CD60a) and 9-O-acetylated GD3 (CD60b). HAP1 cells lack GD3 but can generate the disialo ganglioside after expression of the sialyltransferase ST8Sia I (Fig. 6a; mAb R24). ST8Sia I converts GM3 to GD3 by adding an a2,8-linked Sia residue, which in turn can carry an O-acetyl group in C9 position (Fig. 6c) 49 . In HAP1 wild-type cells, expression of ST8Sia I resulted in the formation of both CD60a and CD60b antigens, as shown by additional staining with mAb UM4D4 (Fig. 6a). As described for other anti-CD60b antibodies, UM4D4 recognizes 9-O-acetylated GD3 and similar gangliosides possessing a terminal 9-O-acetylated disialo group such as 9-Oacetylated disialo neolactotetraosylceramide and 9-O-acetylated disialo neolactohexaosylceramide (see Supplementary Fig. 8c for details on ganglioside structures) 50 . However, binding of mAb UM4D4 strictly relies on the O-acetylation of the carbohydrate epitope and no cross-reactivity towards the unmodified epitope is seen (see refs 50,51 and Supplementary Fig. 8b). Newly synthesized CD60a and CD60b epitopes were both detected at the cell surface, as demonstrated by respective signals on nonpermeabilized cells. In HAP1 DCASD1 cells, expression of ST8Sia I alone led to the formation of GD3 ( Fig. 6b; mAb R24) but not 9-O-Ac-GD3 or other 9-O-acetylated gangliosides of the neolacto series ( Fig. 6b; mAb UM4D4). Ganglioside modification was only detected after co-expression of ST8Sia I with CASD1, demonstrating that CASD1 is crucial for the formation of 9-Oacetylated CD60b epitopes ( Fig. 6b; middle images). Again, alanine exchange of the active site residue S94 drastically impaired CASD1 function ( Fig. 6b; right images). The critical role of CASD1 for the formation of 9-O-acetylated CD60b epitopes was further confirmed in HEK293T wt and HEK293T DCASD1 cells by immunofluorescence and thin-layer chromatography (TLC) analysis, whereas no indication for the presence of 7-O-acetylated disialo gangliosides was found ( Supplementary Fig. 8a,b).

Discussion
Here we define CASD1 as the key factor for Sia 9-O-acetylation, the most common Sia modification in humans. Pivotal to our study was the observation that mammalian cell lines frequently express 9-O-acetylated sialoglycans in the Golgi apparatus, even if 9-O-acetylation of cell surface glycans is non-detectable. This phenomenon, described only by one previous study 37 , highlights that 9-O-acetylation is much more common than anticipated from cell surface staining. The finding that Golgi-confined 9-Oacetylated Sias were attached to N-and O-glycans suggests Golgiresident glycoproteins as underlying carriers and interestingly, 9-O-acetylated Sia was recently found on the N-glycans of a human sialyltransferase 52 .
As shown for HAP1 and HEK293T cells, CASD1 is essential and sufficient for Golgi-confined 9-O-acetylation. However, 9-Oacetylation of cell surface molecules such as GD3 and similar (e) Complementation of HAP1 DCASD1 with CASD1 cDNA. HAP1 DCASD1 cells were transiently transfected with empty vector (mock) or a plasmid encoding epitope-tagged CASD1 ( þ CASD1). Triton-permeabilized cells were co-stained with BCoV-HE 0 -Fc (red) and anti-V5 mAb (green). Nuclei were counterstained with DAPI (blue). Scale bar, 10 mm. DAPI, 4,6-diamidino-2-phenylindole. NATURE COMMUNICATIONS | DOI: 10.1038/ncomms8673 ARTICLE disialo gangliosides required the additional expression of ST8Sia I. This indicates that parental HAP1 cells lack detectable cell surface 9-O-acetylation, because cell surface sialoglycotopes amenable to 9-O-acetylation are either absent or present in very low quantities. Regulation of cell surface 9-O-acetylation by the action of specific sialyltransferases is consistent with data from CHO cells, showing that cell surface 9-O-acetylation could be induced by ST8Sia I or ST6Gal I, two sialyltransferases that are normally not expressed in CHO cells 53 .
Our in vitro data establish that sCASD1, a soluble CASD1 variant encompassing only the SGNH-like domain, transfers acetyl groups from acetyl-CoA to CMP-Neu5Ac, the donor substrate of all sialyltransferases. Although the situation may be more complex in vivo, this result suggests that CASD1 may not directly 9-O-acetylate sialoglycoconjugates, but instead provides 9-O-acetylated CMP-Neu5Ac (CMP-Neu5,9Ac 2 ) for sialyltransferases. In this scenario, the sialyltransferases would play a critical role in the biosynthesis of glycoconjugates that carry 9-Oacetylated Sia. In support of this, CMP-Neu5,9Ac 2 was detected in bovine submandibular glands 54 and biochemical data obtained with human colon mucosa showed a Golgi-associated SOAT activity that utilized CMP-Neu5Ac as acceptor and a sialyltransferase activity that transferred 9-O-acetylated Sia from CMP-Neu5,9Ac 2 to endogenous and exogenous acceptors 33,55 . Moreover, the use of CMP-Neu5,9Ac 2 has been shown for several sialyltransferases and is exploited for chemo-enzymatic synthesis of 9-O-acetylated sialosides 56 . In vivo, the efficient and selective usage of CMP-Neu5,9Ac 2 may rely on a physical interaction between CASD1 and specific sialyltransferases.
Although our data demonstrates that CASD1 mediates 9-Oacetylation, we cannot rule out the possibility that the enzyme initially transfers the acetyl group to the C7 position of Sia followed by migration to position C9 in a non-enzymatic step. Even under physiological conditions, O-acetyl groups at the Sia side chain can spontaneously migrate from C7 and C8 to an unsubstituted hydroxyl group at position C9, where the O-acetyl group is more stable 57,58 . Moreover, 7-O-Ac-specific SOAT activity has been detected in several tissues 31,59,60 and studies on microsomal fractions from bovine submandibular glands provided evidence for an enzyme that accelerates the migration of O-acetyl groups from C7 to C9 in vivo 59 . In line with previous studies on 7-O-Ac-specific SOAT activity, Arming et al. 35 observed a role of CASD1 in 7-O-acetylation but not in 9-Oacetylation of GD3 as shown by FACS analysis with anti-CD60c and anti-CD60b mAbs 35 . The stability of O-acetyl groups at C7 might thus be cell-type-specific and could explain the discrepancy with our data.
By identification of a covalent acetyl-enzyme intermediate that involves the active site residue S94, we provide first insight into the catalytic mechanism of CASD1. Transfer of an acetyl group from acetyl-CoA to a serine placed in a Ser-His-Asp catalytic triad is reminiscent of what we found for the polysialic acidspecific O-acetyltransferase OatC of Neisseria meningitidis 61 and suggests a ping-pong mechanism as described for other members of the SGNH-hydrolase family 62,63 . Consistently, we propose a mechanism in which the catalytic triad of CASD1 forms a charge relay network to activate S94, allowing nucleophilic attack of the carbonyl carbon of the acetyl donor. This leads to the transfer of Siaα2-8Siaα2-3Galβ1-4Glcβ1-1′ Cer (GD3, CD60a) the acetyl group to S94, followed by transfer to CMP-Neu5Ac during the second half reaction. Alanine exchange of S94 in sCASD1 prevented the formation of the acetyl-enzyme intermediate and resulted in a complete loss of SOAT activity in vitro.

9-O -
The soluble sCASD1 variant used for in vitro experiments encompassed only the SGNH-like domain, whereas in vivo, this domain is fused to a C-terminal multi-TM domain. Detection of residual virolectin staining in cells expressing full-length CASD1-S94A may indicate that the C-terminal part contributes to CASD1 activity in vivo. First indication for this scenario is given by a bioinformatics analysis of the cryptococcal Cas1 protein (Cas1p) 39 . This study highlighted that the multi-TM domain found in Cas1p and its animal orthologue CASD1 shares homology with bacterial multi-TM O-acetyltransferases that are involved in the modification of cell surface or periplasmic biopolymers such as polysaccharides and peptidoglycan 39 . Of note, several of the bacterial multi-TM proteins work in concert with an SGNH-like O-acetyltransferase, with the latter being expressed either as separate protein or fused to the multi-TM part as in CASD1 (refs 62,64,65).
In conclusion, we characterized human CASD1 as key enzyme for Sia 9-O-acetylation and provided first insight into the catalytic mechanism. This opens new experimental avenues for the development of effective therapeutic strategies targeting CASD1 to regulate pathological changes in the 9-O-acetylation status and to combat drug-resistant cancer cells in ALL, whose survival crucially depend on 9-O-acetylation 23 .

Methods
Homology modelling. The three-dimensional structure of the catalytic domain of CASD1 was modelled using the Phyre2 protein fold recognition server 66 . On the basis of the structure of the isoamyl acetate-hydrolysing esterase from Saccharomyces cerevisiae (PDB accession code 3mil), which was identified as top rank template, residues 83-290 of human CASD1 were modelled with 99.3% confidence.
For expression of the human sialyltransferase ST8Sia I, a full-length construct encoding an N-terminal V5 epitope was generated by amplification of the coding region of human ST8SIA1 (I.M.A.G.E. clone IRCMp5012B0613D, ImaGenes) using the primers 5 0 -GCTAAGCTTCGAATGGGTAAGCCTATCCCTAACCCTCTCC TCGGTCTCGATTCTACGGGTACCAGCCCCTGCGGGCGGGC-3 0 and 5 0 -GCT GCGGCCGCCTAGGAAGTGGGCTGGAGTG-3 0 containing the sequence encoding the V5 epitope (bold), HindIII and NotI restriction sites (underlined). The resulting PCR fragment was ligated into the HindIII and NotI sites of the expression vector pcDNA3.1-zeo (Invitrogen).
To generate constructs encoding a soluble Fc-chimera of BCoV-HE, DNA encoding the ectodomain (residues 1-388) of BCoV-HE (accession no. AAL57307) was generated by gene synthesis (Eurofins MWG Operon) and subsequent amplification by PCR using the primers 5 0 -GACTGCTAGCATGTTTTTGCTTC CTAGATTTG-3 0 and 5 0 -GACTGGATCCACTTACCTGTATCATACACACA AATAGGTAC-3 0 containing NheI and BamHI restriction sites (underlined), respectively. The obtained PCR product was ligated into the NheI and BamHI sites of pcDNA3.1-zeo, upstream of the sequence encoding the Fc part of human IgG1. A corresponding plasmid that encodes BCoV-HE 0 -Fc harbouring an inactive esterase domain due to the amino-acid exchange S40A was generated by sitedirected mutagenesis using the QuikChange site-directed mutagenesis kit (Stratagene) and the mutagenesis primers 5 0 -GGAGATTGGTTTTTATTTGGTG ACGCCCGTTCAGATTG-3 0 and 5 0 -CAATCTGAACGGGCGTCACCAAATAA AAACCAATCTCC-3 0 . The identity of all constructs was confirmed by sequencing.
Mammalian cell lines and growth conditions. The murine fibroblast cell line LM-TK À (ATCC CCL1.3), the human rhabdomyosarcoma cell line TE671, HEK293T cells and CHO wild type (CHO-K1), CHO lec1, and CHO 6B2 cells were maintained in DMEM/Ham's F-12 1:1 (Biochrom) supplemented with 1 mM sodium pyruvate and 10% fetal calf serum (FCS) at 37°C and 5% CO 2 . Madin Darby canine kidney I cells (MDCK I) were cultured in DMEM (containing 450 mg l À 1 glucose; Biochrom) supplemented with 10% FCS at 37°C and 5% CO 2 . Additional information on the used cell lines are given in the legends to Supplementary Figs 4 and 5. The near-haploid human cell line HAP1 was purchased from Haplogen Genomics (Austria) and maintained in Iscove's modified Dulbecco's medium (Gibco) supplemented with 5% FCS at 37°C and 5% CO 2 . Hybridoma cells producing mAb R24 (ATCC HB-8445) were maintained in RPMI 1640 (Sigma) adjusted to contain 4.5 g l À 1 glucose, 10 mM Hepes, 1 mM sodium pyruvate and 10% FCS. All cell lines were routinely screened for mycoplasma contamination using the PlasmoTest detection kit (InvivoGen).
Inhibition of mucin-type O-glycosylation was achieved by cultivating CHO wild-type and CHO lec1 cells for 3 weeks in medium supplemented with 2 mM benzyl-N-acetyl-a-galactosaminide (benzyl-a-GalNAc; Calbiochem).
Transfection of mammalian cells. Transfection of LM-TK À and CHO cells was performed with Lipofectamine (Invitrogen). Briefly, cells were grown in six-well plates (70-80% confluency), washed twice with PBS and transfected with 1 mg of plasmid DNA and 6 ml Lipofectamine in 1 ml of OptiMEM (Gibco). After 7 h at 37°C and 5% CO 2 the transfection was stopped by adding 1 ml of culture medium supplemented with 10% FCS. For HAP1 and HEK293T cells, the protocol was varied by using 2 ml TurboFectin 8.0 (OriGene Technologies) or 2 ml polyethylenimine 'Max' (Polyscience) instead of Lipofectamine. For the selection of stable transfectants, 750 mg ml À 1 G418 (Calbiochem) or 750 mg ml À 1 zeocin (Life Technologies) were added 72 h after start of the transfection. Colonies were picked and cloned by limiting dilution.
Reverse transcription-PCR analysis. RNA was prepared from mammalian cell lines using TRIzol reagent (Invitrogen) according to the manufacturer's instructions. cDNA was generated from 1 mg of RNA using Revert Aid premium reverse transcriptase (Fermentas) and random hexamer primers according to the manufacturer's recommendations. Gene-specific cDNA was amplified using the following primer pairs: 5 0 -GTTTGGCAACCTCACAGTTGTATGATG-3 0 and 5 0 -AGTTTCTCTGCTCGATTCAG-3 0 (CASD1), and 5 0 -CACACCTTCTACAATG AGCTG-3 0 and 5 0 -GTCAGGTCCCGGCCAGCCAG-3 0 (-ACTIN). PCR products were separated on a 2.5% agarose gel and stained with ethidium bromide. To confirm the presence of a 16-bp microdeletion in CASD1 transcripts of HAP1 DCASD1 cells, CASD1-specific cDNA was amplified using the primer pair 5 0 -CGCCTCCCGCCGCTACCGAGG-3 0 and 5 0 -CAATTGACGAATTCTGGAA TCTC-3, which allows the amplification of a 184-and 168-bp PCR product from wild-type and mutated CASD1, respectively. The obtained PCR products were separated on a 6% polyacrylamide gel and stained with ethidium bromide.
Production and purification of BCoV-HE-Fc and BCoV-HE 0 -Fc. BCoV-HE 0 -Fc and BCoV-HE-Fc were produced by transient expression in HEK293T cells and purified from the tissue culture supernatant as described 42 . As an alternative source of recombinant proteins, CHO cells stably transfected with plasmids encoding BCoV-HE-Fc or BCoV-HE 0 -Fc were cultivated in DMEM/Ham's F-12 1:1 (Biochrom) containing 1 mM sodium pyruvate and 5% low IgG FCS (Gibco).
Tissue culture supernatants from transfected HEK293T cells and conditioned medium of the CHO cells were collected and adsorbed to Protein A Sepharose (GE Healthcare). After washing with 20 column volumes of 20 mM sodium phosphate pH 7.0, bound protein was eluted with 0.1 M glycine-HCl pH 2.8 and neutralized by 1 M Tris-HCl pH 8.5. After addition of NaCl to a final concentration of 150 mM, purified Fc chimeras were stored at -80°C.
Spodoptera frugiperda (Sf9) cells (Gibco) were grown at 27°C in shaking cultures at 80 r.p.m. in protein-free Insect-Xpress medium (Lonza) and maintained at a density of 0.5 Â 10 6 to 6 Â 10 6 viable cells per ml. Transient expression of sCASD1-wt and sCASD1-S94A was performed on infection with recombinant baculovirus particles generated by the Bac-to-Bac system (Invitrogen) and the pFastBac1-based plasmids described above. Conditioned medium was collected 96 h after infection, concentrated 10-fold by tangential flow ultrafiltration (Ultrasette, cutoff of 10 kDa, Pall Corporation), diafiltered against 50 mM sodium phosphate buffer pH 7.5 containing 100 mM NaCl and passed over a HisTrap HP column (GE Healthcare). After washing with 20 column volumes of 50 mM sodium phosphate pH 7.5 containing 100 mM NaCl and 10% glycerol, bound protein was eluted by an imidazole gradient. The imidazole was subsequently removed by gel filtration using a Superdex 200 10/300 GL column (GE Healthcare) equilibrated with 20 mM MES pH 6.8 containing 100 mM NaCl.
Enzymatic removal of N-glycans. To remove N-glycans, 2 mg of sCASD1-wt were incubated with two units of PNGaseF (Alexis Biochemicals) in a total volume of 30 ml of 10 mM Tris-HCl pH 8.0, 1 mM EDTA. Digests were performed at 37°C for the indicated time points.
SDS-PAGE and immunoblotting. Proteins were separated by SDSpolyacrylamide gel electrophoresis under reducing conditions and proteins were either stained with Coomassie Brilliant Blue (Roth) or transferred to polyvinylidene difluoride membranes (GE Healthcare) for western blot analysis. Myc-tagged proteins were detected using mAb 9E10 (5 mg ml À 1 , Roche), followed by goat anti-mouse IgG horseradish peroxidase (HRP) conjugate (1:20,000; Jackson Immunoresearch) and enhanced chemiluminescence detection.
In vitro esterase activity assay. Esterase activity was measured in a total volume of 100 ml containing 15 pmol protein (BCoV-HE-Fc, sCASD1-wt or BSA), 1 mM pNP-acetate (Sigma Aldrich), 50 mM sodium phosphate pH 6.5 and 50 mM KCl. Samples were incubated for 30 min at RT and the absorbance of released pNP was measured at 405 nm.
Radioactive incorporation assay. For radioactive incorporation assays, 5 mg of purified protein (sCASD1-wt or sCASD1-S94A) was incubated for 15 min at 37°C in a total volume of 50 ml containing 50 mM phosphate buffer pH 6.5, 50 mM KCl and 2.5 mM [ 3 H]acetyl-CoA (0.37 kBq ml À 1 , American Radiolabeled Chemicals). To stabilize the acetyl-enzyme intermediate, reaction mixtures were kept at -20°C for at least 18 h. After thawing at RT, an aliquot of 10 ml was spotted on Whatman grade 3MM chromatography paper and free radioactivity was removed by descending paper chromatography using 300 mM ammonium acetate pH 7.5/70% ethanol as mobile phase. Radioactivity incorporated into protein remained at the origin and was quantified by scintillation counting.
In vitro formation of the acetyl-enzyme intermediate. Purified sCASD1-wt or sCASD1-S94A (5 mg) was incubated for 15 min at 37°C with or without 2.5 mM acetyl-CoA (Sigma Aldrich) in 50 ml containing 50 mM phosphate buffer pH 6.5 and 50 mM KCl. After storage at -20°C for at least 18 h to stabilize the acetylenzyme intermediate, the reaction mixture was separated by SDS-PAGE and the protein band was excised, digested with trypsin and analysed by LC-ESI-MS.
In-gel digestion of proteins with trypsin. On separation by SDS-PAGE and staining with Coomassie Blue (Roth), protein bands were excised from gels and subjected to tryptic digest. Briefly, the gel pieces were dehydrated with acetonitrile and rehydrated with 100 mM NH 4 HCO 3 containing 10 mM dithiothreitol (Roth). Subsequently, gel pieces were treated with 100 mM NH 4 HCO 3 containing 100 mM iodoacetamide (Sigma). After a second dehydration step with acetonitrile and rehydration with 100 mM NH 4 HCO 3 buffer followed by dehydration with acetonitrile, the dried gel pieces were rehydrated with 20 ng ml À 1 trypsin (Promega) in 50 mM NH 4 HCO 3 buffer and incubated overnight at 37°C. Peptides were extracted with 75% acetonitrile containing 0.1% formic acid and dried in a vacuum centrifuge.
DMB-HPLC analysis of Golgi-confined Sia in HAP1 cells. For each cell type, HAP1 wt and HAP1 DCASD1 , 2 Â 10 7 cells were incubated in 100 ml PBS containing 15 mU sialidase isolated from Arthrobacter ureafaciens (EY Laboratories, Inc.) for 3 h at 37°C to remove cell surface Sia. Cells were pelleted by centrifugation and lysed in 100 ml of PBS in a Precellys 24 homogenizer (Bertin Technologies) at 5,000 r.p.m. for 3 Â 5 s. The microsomal fraction was obtained by centrifugation for 1 h at 100,000g and removal of the cytosolic fraction. The pellet was washed with PBS followed by centrifugation for 1 h at 100,000g, resuspended in 100 ml of 2 M propionic acid and incubated for 1 h at 80°C to release bound Sia. After centrifugation for 1 h at 100,000g, the supernatant was lyophilized, derivatized with DMB and analysed by HPLC as described.
LC-ESI-MS. MS analysis of peptides and DMB-labelled sugars was performed on a nanoACQUITY UPLC system (Waters) equipped with an analytical column (Waters, BEH130C18, 100 mm Â 100 mm, 1.7-mm particle size) coupled online with an ESI Q-TOF (Q-TOF Ultima, Waters). DMB-labelled sugars dissolved in acetonitrile/methanol/H 2 O (9:7:84, v/v) and peptides dissolved in 2% acetonitrile and 0.1% formic acid were separated by reverse-phase chromatography using acetonitrile as eluent. MS spectra were recorded in positive reflection mode and analytes were automatically subjected to fragmentation (MS/MS). Spectra were analysed using MassLynx V4.1 software (Waters). MS/MS protein spectra were automatically analysed using the program ProteinLynx Global Server (Version 2.1, Waters).