Distinct human α(1,3)-fucosyltransferases drive Lewis-X/sialyl Lewis-X assembly in human cells

In humans, six α(1,3)-fucosyltransferases (α(1,3)-FTs: FT3/FT4/FT5/FT6/FT7/FT9) reportedly fucosylate terminal lactosaminyl glycans yielding Lewis-X (LeX; CD15) and/or sialyl Lewis-X (sLeX; CD15s), structures that play key functions in cell migration, development, and immunity. Prior studies analyzing α(1,3)-FT specificities utilized either purified and/or recombinant enzymes to modify synthetic substrates under nonphysiological reaction conditions or molecular biology approaches wherein α(1,3)-FTs were expressed in mammalian cell lines, notably excluding investigations using primary human cells. Accordingly, although significant insights into α(1,3)-FT catalytic properties have been obtained, uncertainty persists regarding their human LeX/sLeX biosynthetic range across various glycoconjugates. Here, we undertook a comprehensive evaluation of the lactosaminyl product specificities of intracellularly expressed α(1,3)-FTs using a clinically relevant primary human cell type, mesenchymal stem cells. Cells were transfected with modified mRNA encoding each human α(1,3)-FT, and the resultant α(1,3)-fucosylated lactosaminyl glycoconjugates were analyzed using a combination of flow cytometry and MS. The data show that biosynthesis of sLeX is driven by FTs-3, -5, -6, and -7, with FT6 and FT7 having highest potency. FT4 and FT9 dominantly biosynthesize LeX, and, among all FTs, FT6 holds a unique capacity in creating sLeX and LeX determinants across protein and lipid glycoconjugates. Surprisingly, FT4 does not generate sLeX on glycolipids, and neither FT4, FT6, nor FT9 synthesizes the internally fucosylated sialyllactosamine VIM-2 (CD65s). These results unveil the relevant human lactosaminyl glycans created by human α(1,3)-FTs, providing novel insights on how these isoenzymes stereoselectively shape biosynthesis of vital glycoconjugates, thereby biochemically programming human cell migration and tuning human immunologic and developmental processes.

In humans, six ␣(1,3)-fucosyltransferases (␣(1,3)-FTs: FT3/ FT4/FT5/FT6/FT7/FT9) reportedly fucosylate terminal lactosaminyl glycans yielding Lewis-X (Le X ; CD15) and/or sialyl Lewis-X (sLe X ; CD15s), structures that play key functions in cell migration, development, and immunity. Prior studies analyzing ␣(1,3)-FT specificities utilized either purified and/or recombinant enzymes to modify synthetic substrates under nonphysiological reaction conditions or molecular biology approaches wherein ␣(1,3)-FTs were expressed in mammalian cell lines, notably excluding investigations using primary human cells. Accordingly, although significant insights into ␣(1,3)-FT catalytic properties have been obtained, uncertainty persists regarding their human Le X /sLe X biosynthetic range across various glycoconjugates. Here, we undertook a comprehensive evaluation of the lactosaminyl product specificities of intracellularly expressed ␣(1,3)-FTs using a clinically relevant primary human cell type, mesenchymal stem cells. Cells were transfected with modified mRNA encoding each human ␣(1,3)-FT, and the resultant ␣(1,3)-fucosylated lactosaminyl glycoconjugates were analyzed using a combination of flow cytometry and MS. The data show that biosynthesis of sLe X is driven by FTs-3, -5, -6, and -7, with FT6 and FT7 having high-est potency. FT4 and FT9 dominantly biosynthesize Le X , and, among all FTs, FT6 holds a unique capacity in creating sLe X and Le X determinants across protein and lipid glycoconjugates. Surprisingly, FT4 does not generate sLe X on glycolipids, and neither FT4, FT6, nor FT9 synthesizes the internally fucosylated sialyllactosamine VIM-2 (CD65s). These results unveil the relevant human lactosaminyl glycans created by human ␣(1,3)-FTs, providing novel insights on how these isoenzymes stereoselectively shape biosynthesis of vital glycoconjugates, thereby biochemically programming human cell migration and tuning human immunologic and developmental processes.
Display of Le X and sLe X are each very tightly regulated among mammalian cells (1), indicating that they each serve highly specialized biology. Le X is well-known to mediate a variety of important cellular functions in development and immunity. In mice, Le X is known as stage-specific embryonic antigen-1 (SSEA-1); it serves as a major marker of murine (but not human) embryonic stem cells (2,3), and its expression is necessary for compaction of the morula (4). Importantly, in both mice and humans, Le X is a marker for neural stem cells (5)(6)(7)(8), and Le X -bearing glycoconjugates mediate neural stem cell proliferation by activating the Notch signaling pathway (9). Le X is immunomodulatory, serving as one of the main glycans recognized by DC-SIGN (CD209), a C-type lectin (i.e. requiring Ca 2ϩ for ligand binding) expressed by dendritic cells (10). Conspicuously, human (but not mouse) myeloid leukocytes express Le X , and its expression in hematopoiesis is a hallmark of myeloid-specific lineage differentiation. Moreover, Le X is characteristically expressed on Reed-Sternberg cells in Hodgkin's lymphoma (11), and is displayed on certain human vascular and central nervous system malignancies (e.g. gliomas) in which it is considered an indicator of cancer stem cells (12,13).
Although expression of Le X has garnered significant scientific interest, even more attention has been directed to sLe X as this glycan is the prototypical binding determinant for a family of C-type lectins called "selectins" that includes the endothelial molecule known as E-selectin (CD62E) (14). Binding of E-selectin to sLe X -bearing glycoconjugates on circulating cells is critical to enable the deceleration of the flowing cells onto the endothelial surface, which is the key first step in cell migration. In all mammals, E-selectin is constitutively expressed in microvessels in the bone marrow and skin and is inducibly expressed in endothelial beds at inflammatory sites in response to the cytokines tumor necrosis factor and interleukin-1 (15). Expression of cell-surface sLe X is therefore a prerequisite for extravasation of all mammalian leukocytes and for migration of mammalian hematopoietic stem/progenitor cells to marrow (15)(16)(17). Importantly, whereas sLe X plays a critical role in controlling leukocyte and hematopoietic stem/progenitor cell migration, aberrant expression of this tetrasaccharide by human malignant cells is a critical mediator of cancer metastasis (18 -20). In addition to sLe X , E-selectin can also bind to other ␣(1,3)-fucosylated sialyllactosamines known as "VIM-2" (also known as CD65s) (21,22) Fig. S1) (21).
In light of the conspicuously restricted cell expression patterns and the vital roles of Le X and sLe X in human cell biology, it is important to understand how the various ␣(1,3)-FTs shape the intracellular biosynthesis of these glycan determinants in humans. Extensive efforts in the 1980s and 1990s led to the identification and molecular cloning of all six human ␣(1,3)-FUTs that can catalyze creation of Le X and/or sLe X (FUT3 (23), FUT4 (24,25), FUT5 (26), FUT6 (27), FUT7 (28), and FUT9 (29)). These studies functionally characterized the enzymes as isolated proteins, analyzing their biosynthetic properties under non-physiologic ex vivo reaction conditions. Further studies employed a variety of molecular biology approaches in which cell-surface fucosylated glycans were evaluated after transfecting various mammalian cell lines with individual ␣(1,3)-FUT genes (30,31), in some cases in combination with biochemical approaches whereby ␣(1,3)-FT activity was measured in the lysates of such cells using synthetic oligosaccharide (32)(33)(34) and glycolipid (35,36) acceptors. This body of work provided key catalytic insights and yielded working classifications of the product specificities of the ␣(1,3)-FUT gene family, specificities that have been generally recognized as fact for the past quarter century (1, 37) (see Table 1).
There remain two overarching caveats regarding our current understanding of the human ␣(1,3)-FT isoenzymes that modify terminal lactosaminyl glycans: (i) their ex vivo chemistry may not reflect their Golgi chemical biology; and (ii) their chemical biology within immortal cell lines, and, in particular, immortal non-human cell lines, can differ significantly from that within human primary cells. To ascertain how these enzymes contribute to synthesis of Le X and sLe X in context of human cell biology, in this study we systematically evaluated the capability of all six human ␣(1,3)-FTs to generate sLe X and Le X , as well as the related structures VIM-2 and difucosyl sLe X , in primary cultures of human mesenchymal stem cells (hMSCs). hMSCs are an ideal model cell system to study the products of the human ␣(1,3)-FT isoenzymes because they are cultured as primary cells, have immense clinical significance (e.g. they are precursors of bone-forming osteoblasts and have potent immunomodulatory properties), and they can be readily transfected using Lipofectamine-based reagents (38,39). Importantly, these cells are known to natively display sialylated terminal Type 2 lactosaminyl glycans (thus possessing glycosyltransferases necessary to create this structure), but are natively deficient in Le X and sLe X expression (38,40). Accordingly, we transfected hMSCs with modified-mRNA encoding the relevant human ␣(1,3)-FTs (i.e. FT3, FT4, FT5, FT6, FT7, and FT9) to generate a controlled pulse of the pertinent FT activity in these cells. Across multiple donor hMSCs, the resulting cellsurface glycan determinants were examined using the complementary approaches of flow cytometry and mass spectrometry (MS) analysis. Our findings provide new insights on the biosynthetic properties of this important class of glycosyltransferases, yielding novel perspectives on how these enzymes shape expression of functionally significant ␣(1,3)-fucosylated glycan moieties on both glycoproteins and glycolipids in human cell biology.

Lewis-X/sialyl Lewis-X assembly in human cells
cell type would serve as a fitting model system to assess the in vivo substrate specificities of the human ␣(1,3)-FT isoenzymes.

Construction of modRNA encoding the six human ␣(1,3)-fucosyltransferases
To analyze the enzymatic product specificities of the human ␣(1,3)-FTs in the context of primary human cells, we utilized modified-mRNA (modRNA), which is non-permanent and nongenome integrative, and can be readily transfected directly into hMSCs (38). Once inside the cell, the modRNA is recognized by the cell's endogenous translational machinery, and protein synthesis ensues. The resulting protein then undergoes normal post-translational folding and processing, and subsequently becomes localized to its normal position within the cell. We synthesized modRNAs encoding the six human ␣(1,3)-fucosyltransferases (i.e. creating modRNAs FUT3, -4, -5, -6, -7, and -9), transfected them individually into hMSCs, and probed for the presence of surface Le X and/or sLe X by flow cytometry. In initial experiments, the ␣(1,3)-FUT modRNAs generated appreciable levels of Le X and/or sLe X with the exception of FUT4, which displayed little or no activity; despite confirming sequence integrity as previously reported in GenBank TM (accession number: BC136374), and controlling for modRNA integrity and transfection efficiency, we consistently observed no evidence of fucosylated products using multiple independent FUT4 modRNA preparations.

FUT4 modRNA transfection provides direct evidence that only "short" FUT4 RNA creates functional FT4
The original report of cloning of human FT4 indicated that two transcript variants of FUT4 (originally named "ELAM-1 ligand fucosyltransferase" (ELFT)) exist: ELFT (indicated here as "FUT4 SHORT ," which encodes a 405-amino acid protein) and a putative longer isoform (ELFT-L; indicated here as "FUT4 LONG ") (25). Although the molecular basis of these two transcript sizes has not been defined (i.e. alternative splicing of the primary longer transcript or an alternative transcription initiation site embedded within the gene sequence 5Ј to the "short" ELFT gene product), it is recognized that, compared with ELFT, ELFT-L contains an extended 3Ј UTR, a different 5Ј UTR, as well as an alternative ATG start site yielding a coding sequence corresponding to 530 amino acids (Fig. 2). However, the putative ATG start site for ELFT-L is not embedded among any of the nucleotides typical of the consensus Kozak sequence, casting doubt on the translational capacity of this longer RNA transcript (25). Despite this fact, genome databases (e.g. Gen-Bank accession number BC136374), have consistently annotated ELFT-L as the authentic FUT4 mRNA coding sequence, and thus this sequence was chosen for all initial experiments. However, as noted above, hMSC transfection with FUT4 LONG -modRNA did not yield fucosylated products on the cell surface, and we thus tested whether transfection with FUT4 SHORT -modRNA could yield enzymatic function. To this end, we performed transfection in hMSCs using FUT4 SHORT -modRNA versus FUT4 LONG -modRNA, and, as can be seen in Fig. 2, FUT4 SHORT -modRNA produced robust levels of cell-surface Le X , whereas FUT4 LONG -modRNA did not produce Le X . Although this variability of the human FUT4 RNA is not well-recognized, our findings corroborate the results of the initial human FT4 cloning report that suggested that ELFT (FUT4 SHORT ) encodes functional FT4 (25).
Because the qRT-PCR primer sets used initially to probe FUT4 transcripts in hMSCs were embedded within the "common" FUT4 coding sequence (i.e. amplifying a 123-bp fragment within the coding region common to both FUT4 LONG and FUT4 SHORT ) (Table S1), we analyzed for FUT4 LONG (i.e. ELFT-L) transcripts using a 5Ј primer that detects a 150-bp amplicon upstream of (and not overlapping with) the core FUT4 sequence (Table S1). Plasmid containing FUT4 LONG cDNA was used as positive control, as was RNA isolated from the human promyelocytic cell line HL60 (which is known to express both FUT4 transcripts) (25,44). qRT-PCR of HL60 cells confirmed the presence of the FUT4 LONG transcript and the common FUT4 transcript (Fig. 3A), with amplicon sizes verified by gel electrophoresis of the PCR products (Fig. 3B). In hMSCs, a low transcript level was detected with FUT4 commonprimers, whereas no product was detected with FUT4 LONGprimers (Fig. 3A). These findings indicate that only the short form of FUT4 is expressed in hMSCs, although neither Le X nor sLe X product was detectable on the cell surface by flow cytometry.

Lewis-X/sialyl Lewis-X assembly in human cells modRNA-driven expression of ␣(1,3)-FTs reveal unique product specificities
We transfected hMSCs derived from multiple marrow donors with each of the ␣(1,3)-FUTs (n ϭ 15 donors, for FUT-3, -5, -6, -7, -9; n ϭ 9 donors, for FUT4 SHORT ), and 48 h later measured the resulting fucosylated lactosaminyl glycans using monoclonal antibodies recognizing the terminally ␣(1,3)-fucosylated lactosaminyl glycans Le X and sLe X , and the related internally (penultimate) ␣(1,3)-fucosylated lactosaminyl glycans VIM-2, as well as the determinant containing both terminal and penultimate ␣(1,3)-fucose modifications known as difucosyl sLe X (Fig. 4, Fig. S3). All six of the ␣(1,3)-FTs tested were able to fucosylate Type 2 lactosamines to create Le X and/or sLe X , although clearly distinct enzymatic specificities were observed (Fig. 4). With the exception of FT9, all other ␣(1,3)-FTs were able to terminally fucosylate a sialylated Type 2 lactosamine, with FT6 and FT7 having the highest potency, FT3 and FT5 having moderate capability, and FT4 showing very limited capacity to create sLe X (Fig. 4A, Fig. S3A). Conversely, FT9 and FT4 were most potent in terminally fucosylating an unsialylated Type 2 lactosamine to create Le X . FT3, FT5, and FT6 had moderate ability, and FT7 did not create Le X (Fig. 4B,  Fig. S3B). Notably, FT3 and FT5 were the only FTs that could internally fucosylate a sialylated polylactosamine, creating the VIM-2 glycan determinant (Fig. 4C, Fig. S3C). FT3, FT5, FT6, and FT7 were all capable of fucosylating both the penultimate and terminal lactosamines of a sialylated poly-lactosamine, thereby creating the difucosyl-sLe X structure (Fig. 4D, Fig.  S3D). Consistent with earlier reports (Table 1), FT4 and FT9 do not create any detectable difucosyl sLe X , but, in contrast to prior findings, FT6 did so with the highest potency of all the The results of flow cytometry studies are summarized in Fig.  6A. Importantly, we observed strict consistency in this expression profile across all hMSC cultures, indicating that this panel identifies a "lactosaminyl glycosignature" of this human cell type. While our findings generally parallel the results of earlier in vitro assays, as well as the data derived from transfection studies of both non-human and human cell lines (Table 1), there are several divergences between our dataset and the prior findings: the VIM-2 glycan is not created by FT4, FT6, or FT9; there was modest but consistently detectable production of sLe X following FUT4 SHORT -modRNA transfection; and, although FT6 does not create VIM-2, it creates significant amounts of difucosyl sLe X .

Lewis-X/sialyl Lewis-X assembly in human cells Glycomic analysis of modRNA-transfected hMSCs
Flow cytometry analysis using monoclonal antibodies provides very useful information regarding the relative amounts of fucosylated lactosaminyl glycan determinants created by ␣(1,3)-FTs but is not able to distinguish their display on different scaffolds, such as O-linked versus N-linked glycans of pro- A, sLe X (measured by antibody CSLEX1) and B, Le X (measured by antibody HI98) levels measured on the cell surface. Non-transfected (NT) hMSCs and matched FUT-3, -5, -6, -7, and -9 -transfected hMSCs were compared relative to each other (n ϭ 15), while FUT4 -transfected hMSCs were compared separately along with a partially overlapping set of matched NT controls (n ϭ 9). Statistics performed were repeated measures of ANOVA (p Ͻ 0.0001 for both A and B) followed by a Tukey post-test, comparing all columns with each other. Symbol above each bar compares to NT. For sLe X , FUT9 is significantly lower than FUT-3, -5, -6, and -7 (p Ͻ 0.001), and for Le X , FUT7 is significantly lower than FUT3, FUT5, FUT6, and FUT9 (p Ͻ 0.001). C, cell-surface VIM-2 levels (paired analysis, n ϭ 5). Statistics performed were repeated measures of ANOVA (p Ͻ 0.001) plus Dunnett's post-test, comparing all modRNA-transfected samples with control (NT). Symbols above each bar compare to NT. D, cell-surface difucosyl sLe X levels were measured by antibody FH6 binding (paired analysis, n ϭ 3). Statistics performed were repeated measures of ANOVA (p Ͻ 0.001) plus Dunnett's post-test, comparing all modRNA-transfected samples with control (NT). Symbol above each bar compares to NT, n.s. ϭ p Ͼ 0.05; *, p Ͻ 0.05; **, p Ͻ 0.01; ***, p Ͻ 0.001. Individual data points are depicted, with error bars representing S.D.

Table 1 Summary of human ␣(1,3)-fucosyltransferase substrate specificities based on current literature
The symbols indicate the ability of the enzyme to create the indicated structure with: ϩ, low; ϩϩ, moderate; and ϩϩϩ, high potency. The -indicates inability to create the indicated structure, and (ϩ) indicates controversy exists regarding product formation.

Lewis-X/sialyl Lewis-X assembly in human cells
tein, or glycosphingolipids (GSLs). However, identification of these various structures can be achieved by glycan fractionation followed by mass spectrometry (MS), and, furthermore, ion trap sequential MS (MS n ) glycan analysis can provide detailed assessment of the component monosaccharide composition; this information complements and cross-validates the molecular identification of the relevant lactosaminyl glycan determinants detected by flow cytometry. To perform MS, cell lysates were prepared from different hMSC cultures derived from multiple donors at 48 h after transfection with individual FUT modRNAs. O-Glycans, N-glycans, and GSL-glycans were sequentially released from lysates and isolated. These fractions were then interrogated by various MS techniques to determine the presence or absence of specific fucosylated structures. Reduced and permethylated N-glycan samples were subjected to electrospray ionization by direct infusion into an ion trap mass spectrometer. Fig. 5 presents a representative mass spectra and sequential MS n spectra of the N-glycans of nontransfected (Fig. 5A) and FUT6-transfected (Fig. 5B) hMSCs. The positive mode mass spectra can only determine the overall glycan compositions in these samples. To assess the synthesis of Le X /sLe X formation in N-glycans, detailed interrogation is required due to the presence of multiple isomeric structures for any sialofucosylated composition. We chose to focus on an N-glycan composition found in all samples, a composition with an m/z 1322 that is doubly-charged, and has the configuration, NeuAc 1 Hex 2 HexNAc 2 dHex 1 ϩ Man 3 GlcNAc 2 , corresponding to a biantennary, monosialylated, and monofucosylated structure. In this ion, the fucose ("dHex") can be positioned either on the core GlcNAc (i.e. the ␣(1,6)-fucosylation catalyzed by FT8), or on the unsialylated LacNAc antenna (i.e. Le X or H type 2 ("H2," the blood group O antigen on a Type 2 lactosamine)), or on the sialylated LacNAc antenna (i.e. sLe X ) (Fig. S4), and, thus, the m/z 1322 2ϩ ions could contain one or more of the following isomeric possibilities: core fucosylation, Le X , the H2 structure, or sLe X . Using ion-trap MS n we could perform detailed structural analysis, and thus identify which of these isomers represent this composition. Specifically, the fucosylated fragment ions of interest include m/z 490 (core GlcNAc fucosylation), m/z 660 (terminal fucosylated LacNAc, i.e. Le x and/or H2), m/z 1021 (sLe X tetrasaccharide), and 646 (a desialylated sLe X , generating an Le X fragment) (Fig. S4). The MS 2 spectra of the m/z 1322 2ϩ ions from nontransfected (Fig. 5C) and FUT6-transfected (Fig. 5D) hMSCs indicate that non-transfected hMSCs contain the core fucosylation fragment, but negligible antennal fucosylation. The FUT6-transfected sample on the other hand contains detectable fucosylated LacNAc fragments, both terminal (m/z 660) and internal (m/z 646), although these ions are much less abundant than the core-fucosylated fragment. MS 2 is not sufficient to determine whether H2 or Le X yields the terminally fucosylated fragment (m/z 660). Thus, MS 3 analysis was done to definitively characterize the m/z 660 ion (Fig. 3E) from the FUT6-transfected hMSCs. This spectrum, as described in an earlier article (45), empirically identifies the presence of Le X in the FUT6-transfected hMSC sample. Fig. 5F shows the MS 4 spectrum of the internal LacNAc fragment (m/z 646). For this analysis, the sialofucosylated tetrasaccharide (m/z 1021) was isolated first, followed by fragmentation into the internal fuc-LacNAc trisaccharide (m/z 646). This spectrum and fragmentation pathway empirically identifies the sLe X motif (45). Identical analyses were done to identify these structures in hMSCs transfected with FUTs-3, -4, -5, -7, and -9. In addition, in an

Lewis-X/sialyl Lewis-X assembly in human cells
effort to assess for the presence of difucosyl sLe X , we selected the m/z 1634 2ϩ ion; this composition is representative of three lactosamine units, one sialic acid, three mannose, two GlcNAc, and two fucose molecules. This composition could be contained within both triantennary and biantennary structures, with the former containing a LacNAc on each antenna and the latter containing one antenna containing a single Lac-NAc and the other containing two contiguous LacNAc units (a "diLacNAc" motif). By selecting the putative sLe X -Le X heptasaccharide fragment (difucosyl sLe X ) and then obtaining an MS 3 spectrum, we could then assess for the presence of this motif (Fig. S4F). The results indicate that core ␣(1,6)-fucosylation is uniformly detectable in hMSCs, whereas sLe X is not detectable among native hMSCs and its creation is unique to specific fucosyltransferases (Fig. 6B). By MS analysis, traces of Le X were observed only on N-glycans of native hMSCs, which may reflect the endogenous low-level expression of FUT4 SHORT transcripts, and, notably, this modest level of Le X expression was not detectable by flow cytometry. In contrast, although VIM-2 determinants were readily identified by flow cytometry in FUT3 and FUT5 transfectants, the MS n fragmentation profiles representative of this structure could not be definitively detected on any glycoconjugate sample from any transfectant. The incongruity between flow cytometry and MS detection of VIM-2 expression is explained by the fact that MS identification of this structure requires isolation of the hexasaccharide fragment with subsequent dissociation, and the lability of linkages in this structure, coupled with its apparent low abundance and the presence of other isobaric ions in these samples, prevented a confident MS assignment.
Because O-glycan fractions generally have fewer sample components than N-glycans and also are generally small-sized glycans, their greater polarity relative to permethylated N-glycans or glycosphingolipids makes reversed-phase LC-MS a useful approach to obtain relative quantitative data. By selecting relevant mass ranges, extracted ion chromatograms can be used to provide data informing both glycan structure and relative quantity. Since VIM-2 and difucosyl sLe X were not detected on O-glycans by direct infusion MS analysis, using LC-MS we focused on ions corresponding to simple branched O-glycan structures containing Le X or sLe X , or the corresponding unfucosylated acceptor structures ( Fig. S6 and S7). By comparing the relative amounts of the product and acceptor, an approximate percent created could be estimated, enabling a semiquantitative measurement of Le X and sLe X creation on O-glycans (Fig. 6, B and C). When Le X and sLe X creation on O-glycans was thus calculated, and compared with the flow cytometry data measured from the same samples, the data were strikingly similar (Fig. 6C). Besides cross-validating the specificity, sensitivity, and qualitative and quantitative utility of both approaches for detection of Le X and sLe X , the close correlation between results from flow cytometry and MS analysis of O-glycans indicates that, for each transfectant, there was no preferential skewing of terminal fucosylation away from O-glycan acceptors. However, a major difference is found in expression of difucosyl sLe X , which is clearly measurable by flow cytometry but MS indicates presence of this structure solely on N-glycans. Importantly, FT6 and FT7 can each construct difucosyl sLe X but not VIM-2; thus, for creation of difucosyl sLe X , it is most likely that FTs-3, -5, -6, and -7 first fucosylate the terminal GlcNAc (thereby creating sLe X ), followed by fucosylation of the internal GlcNAc, preferentially on N-glycans.
For MS analysis of glycosphingolipids, two clusters of peaks were chosen for detailed analysis: m/z 1747, which would contain the fucosyltetraosylceramides (including Le X ), and m/z 1066, which would contain the doubly-charged sialofucosylated tetraosylceramide (including sLe X ) (Fig. S5). MS n was then performed on each of these m/z ranges to identify the specific fucosylated glycans present (if any) for each FUT modRNA transfectant. As summarized in Fig. 6B, Le X was detected in FUT4-, FUT6-, and FUT9-transfected hMSCs, , the relative range for that epitope created by various FTs is: insignificant amounts created on the cell surface (Ϫ); and comparatively low (ϩ), moderate (ϩϩ), or high (ϩϩϩ) levels created on the cell surface. NT, non-transfected. B, summary table of all MS data. Glycans were fractionated into N-glycans, O-glycans, and GSLs and analyzed separately for Le X , sLe X , and difucosyl-sLe X . N-Glycan and GSL analysis was not quantitative and results are presented as present, trace, or undetected (U). O-Glycan analysis was semi-quantitative (Fig. S6), and was classified as (Ϫ) insignificant, (ϩ) low, (ϩϩ) moderate, and (ϩϩϩ) high. C, comparison of Le X and sLe X creation on O-glycans (measured by LC-MS, n ϭ 3 to 6) and on the cell surface (measured by flow cytometry, n ϭ 9 to 15) after transfection with FUT modRNA. Error bars represent S.D.

Lewis-X/sialyl Lewis-X assembly in human cells
which corresponds to the three fucosyltransferases that resulted in the highest levels of Le X by flow cytometry. Notably, despite the ability to produce Le X and sLe X on glycoproteins, FUT5-transfectants did not make any detectable Le X or sLe X on GSLs, and FUT3-transfectants made only a trace of Le X on GSLs, suggesting a unique substrate specificity and/or subcellular localization pattern of FT3 and FT5 compared to other FTs. Importantly, sLe X bearing GSLs were detected prominently only in FUT6-transfected hMSCs, indicating that FT6 is either specialized to create sLe X on GSLs or has unique accessibility to GSLs inside the Golgi.
Collectively, the MS results indicate that, with exception of trace amounts of Le X on N-glycans, all other types of terminally fucosylated LacNAc are not natively expressed on hMSC glycoconjugates. Moreover, both unsialylated and sialylated acceptors are natively distributed among N-glycans, O-glycans, and GSLs in hMSCs. Notably, hMSCs exclusively possess Type 2 lactosamines (Gal␤(1,4)-GlcNAc), as empirical MS n analysis of LacNAcs showed no evidence of Type 1 (Gal␤(1,3)-GlcNAc) structures. Moreover, we assessed production of sialyl Lewis-a (sLe a ) among the various hMSC-FUT transfectants by measuring binding of antibody CA19-9. In sharp contrast to the marked increases observed in HI98-reactivity (Le X ) and CSLEX1-reactivity (sLe X ), the FUT transfectants (particularly as might be expected from the specificities of FT3 and FT5 for Type 1 lactosamine acceptors) did not yield significant CA19-9 reactivity (sLe a ), thus indicating paucity of Type 1 lactosamine acceptors in hMSCs.

Discussion
To understand cellular glycobiology, it is imperative to elucidate the molecular effectors of glycan biosynthesis, the glycosyltransferases. However, elucidating the authentic drivers programming assembly of a particular glycan moiety of interest is challenging because of the following reasons: (i) in contrast to nucleic acids and proteins, glycan construction is not templatedriven, it is directed by the sequential action(s) of glycosyltransferases that modify relevant acceptors in a stereospecific fashion; (ii) glycosyltransferase isoenzymes typically display functional redundancy (i.e. glycosyltransferases commonly create more than one glycan product, and the same glycan structure may be generated by many glycosyltransferases); and (iii) glycosyltransferase expression patterns vary in a species-specific and cell-specific fashion. Thus, to determine whether a certain glycosyltransferase can mediate construction of a given glycan, the relevant enzyme property can be analyzed using ex vivo or in vivo systems. However, because ex vivo studies of glycosyltransferases provide insights only on what these enzymes are capable of producing under the defined reaction conditions utilized, it is imperative to undertake studies in living cells to reveal their true biologic activity.
To date, our understanding of the product specificities of human ␣(1,3)-FTs is largely based on ex vivo studies. Such studies employed purified or recombinant ␣(1,3)-FTs, or extracts of cells expressing these enzymes, to catalyze fucose addition in cell-free reactions targeting synthetic oligosaccharide substrates presented on artificial reaction surfaces, each present at supraphysiologic concentrations. These studies also typically utilized extremely high levels of manganese (Mn 2ϩ ) as a co-factor (i.e. as high as 40 mM, whereas, human physiologic manganese levels are Ͻ20 nM), raising fundamental questions about the relevance of these findings to native human cell biology (46). Thus, although such experimental systems can define the catalytic properties of an isolated enzyme, it cannot be assumed that the data derived therefrom can be extrapolated to cell biology. However, prior studies investigating the in vivo glycobiology of human ␣(1,3)-FTs utilized a wide variety of immortalized mammalian cell lines (predominantly non-human, e.g. CHO, COS, and BHK cells) as hosts for human ␣(1,3)-FUT transfection(s), and evaluated product profiles by flow cytometry of the cell surface. These diverse cell line-based studies provided broad insights that could only tentatively predict the ␣(1,3)-FT product specificities within primary human cells. Thus, the non-physiological systems employed in prior in vitro studies and the studies involving a wide range of cell/animal models have raised questions regarding the translatability of these findings to human cell biology, especially in the context of primary human cells.
In this study, we evaluated the impact of FT3, FT4, FT5, FT6, FT7, and FT9 on expression of Le X and sLe X moieties, as well as the structurally related VIM-2 and difucosyl sLe X determinants, on both protein-based and lipid-based glycoconjugates in primary cultures of hMSCs. Besides these six ␣(1,3)-FTs, two other ␣(1,3)-FTs, known as FT10 and FT11, have been identified. These enzymes were not evaluated as they reportedly modify GlcNAc only within the chitobiose core of N-glycans (47) (a modification not typical of mammalian cells). A recent study in FT9-knockout mice has claimed a possible role for FT10 in creation of Le X on a uniquely restricted set of N-glycan acceptors found only in mouse neural progenitor cells, but, importantly, a contributory role for FT4 was not formally excluded in that model system as FT4/9 double knockout mice were not studied (48).
We employed transfection of modRNA encoding the relevant ␣(1,3)-FT transcripts to enable assessment of the relative "in-Golgi" capacity of these isoenzymes to program expression of ␣(1,3)-fucosylated terminal lactosaminyl glycans. Our results are consistent with prior findings from both ex vivo and cell line-based studies (reviewed in Table 1) with regard to the capacity of the various ␣(1,3)-FTs to create Le X : FT4 and FT9 in hMSCs are most potent at fucosylating unsialylated Type 2 lactosamines, with FT6 demonstrating moderate capability, FT3 and FT5 showing weak ability, and FT7 having little to no ability. However, we observed key differences from historical data regarding the capacity of the various ␣(1,3)-FTs to produce sLe X and, also, VIM-2 and difucosyl sLe X . Similar to earlier studies, we found that FT6 and FT7 have the highest potency in fucosylating sialylated Type 2 lactosamines (thereby creating sLe X ), followed by FT3 and FT5, whereas FT9 is unable to generate any significant levels of sLe X . Notably, whereas prior studies using the human prostate cancer cell line PC-3 (49) and using Chinese hamster ovary (CHO) cell lines (50,51) each indicated that transfection of human FUT4 yields readily detectable levels of sLe X , we observed very limited production of sLe X by FT4 in hMSCs. In terms of VIM-2 expression, although MS was unable to resolve this structure within the

Lewis-X/sialyl Lewis-X assembly in human cells
various hMSC transfectants, flow cytometry analysis using anti-VIM-2 mAb clearly indicated that FT3 and FT5 are the only human ␣(1,3)-FTs that construct this glycan determinant on hMSCs. Again, key differences exist between our results and those of prior cell line-based studies indicating that VIM-2 could be created in CHO cells by FUT4 (24) and FUT6 (35) transfection, in Jurkat cells (a human lymphoblastic cell line) by FUT4 transfection (52), and in Namalwa cells (a human lymphoma cell line) following FUT9 transfection (53). Other cell line studies have also indicated that human FT9 can create the VIM-2 determinant on CHO Lec29 cells (54) and HEK293T cells (a human embryonic kidney line) (55), with commensurate induction of E-selectin binding in each (54,55). Additionally, although FT3 and FT5 were the only enzymes that created VIM-2 in hMSCs, transfection of FUT3, FUT5, FUT6, and FUT7 resulted in cell-surface expression of difucosyl sLe X . The finding that difucosyl sLe X is created in FUT6 and FUT7 transfectants in the absence of measurable VIM-2 production suggests that these enzymes modify a sialylated di-LacNAc by first fucosylating the terminal GlcNAc to create an sLe X structure, then act to convert this structure to difucosyl sLe X via the addition of a fucose to the penultimate GlcNAc. In other words, these enzymes do not appear to make an intermediate VIM-2 structure, because they have a dominant preference for installing a fucose at the terminal GlcNAc. FT3 and FT5 are the only enzymes that can create VIM-2 in hMSCs, and, importantly, appear to have a relatively balanced production of both VIM-2 and sLe X determinants. Collectively, these aforementioned variations in products created by ␣(1,3)-FTs as reported using mammalian cell lines and as we have observed in primary hMSCs can be reconciled by considering Golgi dynamics, including factors such as glycosyltransferase localization within specific Golgi microdomains (56) and/or cell-specific variation(s) in Golgi topography that impact glycan formation. Additionally, the accessibility of acceptor glycans within such microdomains, and intra-Golgi dynamics of competing glycosyltransferases and of glycosidases that can modify the same glycan acceptor, could alone and in combination amend the product profile(s) for any given glycosyltransferase within a given cell. In addition to such variations in Golgi dynamics, it is important to consider how phylogenetic disparities impact our understanding of the glycobiology of the ␣(1,3)-FTs. In this regard, an early study of Fut7 KO mice identified this ␣(1,3)-FT as key to creation of sLe X displayed on selectin ligands, emphasizing its role in regulation of leukocyte homing (57). Subsequently, it was determined that FT4 also had a role in this process, since Fut4/Fut7 double knockout mice yielded more severe leukocyte homing deficits (58 -60). Importantly, while FT4, FT7, and FT9 are expressed in both rodents and primates, three of the six human ␣(1,3)-FUTs (FUT3, FUT5, and FUT6) arose from a (relatively) recent gene duplication event in primates and are not present in the rodent genome (61). Our study shows that FT3, FT5, and FT6, the members of the human ␣(1,3)-FUT gene family without functional homologs in mice, can each produce all three forms of terminally fucosylated lactosamines: Le X , sLe X , and difucosyl sLe X . Among these "primate FTs," the FT3 and FT5 isoenzymes display the broadest product profiles as they can also generate the internally fucosylated VIM-2 epitope. In contrast, FTs-4, -7, and -9 exhibit the most restricted product specificities. Taken together, the stark contrast in ␣(1,3)-FT repertoire between human and murine systems emphasizes the importance of undertaking studies in human cells using human glycosyltransferases to appropriately understand human glycobiology.
Fucosylated terminal lactosaminyl glycans can be displayed on glycoprotein N-glycans or O-glycans or on GSLs. In prior studies of ␣(1,3)-FT product specificities, the carrier scaffold "preferences" were commonly not analyzed, although it is wellrecognized that the scaffolds play important roles in modulating the function of the glycan moiety. For example, E-selectin ligand motifs (i.e. sLe X , VIM-2, and difucosyl sLe X ) on O-glycans and N-glycans participate in initiating E-selectin-mediated tethering and rolling of human myeloid leukocytes (62), however, when presented on glycolipids, they also contribute measurably to stabilizing this interaction and critically facilitate the slow rolling of leukocytes on the endothelial cells (63). Using MS, we could comprehensively delineate the glycoconjugate repertoire of fucosylated terminal lactosaminyl glycans, distinguishing whether Le X , sLe X , and difucosyl sLe X were displayed on glycoprotein N-glycans, O-glycans, or GSLs for each ␣(1,3)-FUT transfectant. Our data indicate that, generally, Le X production by ␣(1,3)-FTs showed no significant preference(s) for one scaffold versus another, with the exception of FT5, which does not create Le X on GSLs (Fig. 6B). However, for sLe X , major variations in scaffold preference were observed: FT6 is the only ␣(1,3)-FT specialized to construct sLe X on all three types of glycoconjugate scaffolds, and, contrary to a prior report indicating that human FUT4 transfection in CHO cells robustly creates sLe X preferentially on glycolipids (51), flow cytometry did not detect any significant level of sLe X expression on hMSC FUT4 transfectants; MS analysis of FT4 products revealed no sLe X on glycolipids and only trace sLe X on O-and N-glycans, whereas FT4 generated Le X robustly on all types of glycoconjugates. Additionally, in hMSCs, difucosyl sLe X was detected solely on N-glycans of FUT3, FUT5, FUT6, and FUT7 transfectants, but others have reported that this structure resides abundantly on GSLs (64).
Our study presents the first systematic approach not only to operationally define the human ␣(1,3)-FTs within the context of a primary cell, but also to elucidate how these various isoenzymes shape lactosaminyl glycan biosynthesis within a given cell type. Importantly, the pattern of glycans and their associated glycoconjugates produced by each FT was uniformly consistent among the numerous marrow-derived hMSC cultures for that distinct FUT modRNA transfectant, indicating that modRNA transfection yields reproducibly functional glycosyltransferases that operate with high fidelity and, additionally, that hMSCs possess a characteristic and well-defined lactosaminyl glycan "glycosignature." The in-depth characterization of product specificities of the individual human ␣(1,3)-FTs provided by our work now offers a useful foundation for designing novel strategies for utilizing these glycosyltransferases, individually or in combination, to custom-engineer fucosylated lactosaminyl glycans for further elucidation of the role(s) of these structures in cell biology and for potential therapeutic purposes. In particular, hMSCs have drawn immense

Lewis-X/sialyl Lewis-X assembly in human cells
attention for clinical applications, as they natively possess considerable immunomodulatory properties, support tissue regeneration, and can themselves differentiate into various cell types essential for normal physiology (e.g. osteoblasts). Notably, although our results identify several differences in the human ␣(1,3)-FT product profiles between previously tested cell lines and hMSCs, many of our results broadly conform to the earlier findings. These disparities are reflective of the differences in the underlying biology of the cell types. Thus, while we recognize that the observed hMSC product profiles of the fucosyltransferases cannot be used as a broad standard per se, our data highlight a very important aspect of glycobiology, which we believe cannot be overstated: one must not overlook host cell biology nor the nonphysiologic nature of in vitro assays in establishing algorithms predicting glycosyltransferase product profiles in vivo.

Isolation and culture of hMSCs
Human cells were obtained and used in accordance with the Declaration of Helsinki and with the procedures approved by Dana Farber/Harvard Cancer Center (DF/HCC) Institutional Review Board (IRB). Bone marrow cells from normal human donors were obtained from discarded bone marrow filter sets by flushing with PBS plus 10 units/ml of heparin (Hospira, Lake Forest, IL). The cells were then gradiented using Ficoll-Histopaque 1.077 (Sigma) and the mononuclear fraction was collected. The cells were suspended in MSC media, consisting of Dulbecco's modified Eagle's medium low glucose (1 g/liter) supplemented with 10% fetal bovine serum from select lots (Atlanta Biologicals, Atlanta, GA), 100 units/ml of penicillin, and 100 units/ml of streptomycin. Approximately 5 ϫ 10 8 cells were seeded into T-175 tissue culture flasks and incubated overnight in a humidified incubator at 37°C, 20% O 2 , 5% CO 2 . The next day, all nonadherent cells were removed and the flasks were rinsed several times with PBS. The adherent cells were cultured in fresh MSC media, with a full media change twice per week. When the cells approached 80% confluence, cells were harvested with 0.05% trypsin, 0.5 mM EDTA, washed with PBS for experimental use and/or passaged by diluting 3-to 5-fold in fresh MSC media. Cells were cultured for a maximum of 6 passages before discarding.

Quantitative RT-PCR
Total cellular RNA was purified from bone marrow-derived hMSCs, using RNeasy microkit (Qiagen, Hilden, Germany) as per the manufacturer's instructions. Total RNA was reverse transcribed to synthesize first strand cDNA using a iScript cDNA synthesis kit (Bio-Rad) with heat cycles as follows: 25°C for 10 min, 42°C for 60 min, and 85°C for 5 min. For some experiments, a SuperScript VILO cDNA conversion kit (Invitrogen) was used for first strand synthesis using identical thermal cycling conditions. Quantitative real time PCR was performed with specific primers to amplify glycosyltransferase genes (Table S1) using the SYBR Select master mix (Applied Biosystems, Foster City, CA) and a StepOne Plus PCR detection system (Applied Biosystems). PCR for individual genes were performed in triplicate with each reaction containing 1% (v/v) of the total cDNA obtained. The following cycling condition was employed: initial activation at 95°C for 20 s, followed by 40 cycles of 95°C for 10 s, and 60°C for 30 s. Post-amplification, melt curve analysis was performed to ensure primer binding specificity according to the following conditions: 95°C for 15 s, 60°C for 1 min, 3°C increment every 15 s to reach 95°C. Gene expression levels were evaluated relative to glyceraldehyde-3phosphate dehydrogenase using comparative C T methods. In some cases, agarose gel electrophoresis was performed with the PCR product to confirm the size of the amplicon. To this end, PCR products were resolved in a 1.3% agarose gel with 0.2-0.5 g/ml of ethidium bromide. DNA fragments were visualized and imaged using UV illumination (Alphaimager EC imaging system, Alpha-Innotech, San Leandro, CA).

Cloning of FUTs
Full-length human fucosyltransferase cDNAs were purchased from Open Biosystems (GenBank accession numbers for FUT-3, -4, -5, -6, -7, and -9 cDNAs are BC108675, BC136374, BC140905, BC061700, BC074746, BC036101, respectively) and amplified by PCR using HiFi Hotstart (KAPA Biosystems, Wilmington, MA). PCR products were then subcloned into the pORFIN plasmid (65), which contains the T7 promoter, 5Ј UTR and 3Ј UTR required for eventual modified mRNA production, using a Quick Ligation Kit (New England Biolabs, Ipswich, MA). The PCR products were first separated using an agarose gel. The products with expected size were gel-purified using QIAquick Gel Extraction Kit (Qiagen) and sequenced to check for mutations before ligation into pORFIN. PCR primer sequences used for cDNA amplification were as follows:

Modified mRNA synthesis
Modified mRNAs (modRNAs) were synthesized as described previously (65). Briefly, human FUT cDNAs cloned in the pORFIN vector were amplified by PCR using HiFi Hotstart (KAPA Biosystems) to generate templates for in vitro transcription. 1.6 g of the PCR products (including the human FUT open reading frames and 5Ј-and 3Ј-untranslated regions) were then used as templates for modRNA synthesis using the MEGAscript T7 kit (Ambion). 3Ј-0-Me-m 7 G(5Ј)ppp(5Ј)G ARCA cap analog (New England Biolabs), adenosine tri-

Lewis-X/sialyl Lewis-X assembly in human cells
phosphate, guanosine triphosphate (U. S. Biochemical Corp.), 5-methylcytidine triphosphate and pseudouridine triphosphate (TriLink Biotechnologies, San Diego, CA) were used for in vitro transcription. Synthesized modRNA product was purified using MEGAclear spin columns (Ambion), and aliquots were stored frozen for future use at Ϫ80°C.

Modified mRNA transfection of hMSCs
modRNAs were transfected into hMSCs using Stemfect (Stemgent, Lexington, MA) as per the manufacturer's instructions. Briefly, 1 g of modRNA and 2 l of Stemfect reagent were individually diluted into Stemfect buffer, mixed together, and incubated for 15 min at room temperature. The mixture was then added to 1 ϫ 10 6 recently harvested hMSCs that were suspended in 2 ml of MSC medium, and incubated at 37°C for 1 h, inverting the tube at 15-min intervals. The cells were then diluted in MSC media supplemented with 200 ng/ml of interferon inhibitor (B18R, eBioscience) and cultured in tissue culture flasks for 2 additional days prior to harvest and downstream analysis.

Glycan release and processing
hMSCs were washed with PBS and cell pellets were frozen until use. Cell pellets were thawed, dried, and transferred to 13 ϫ 100-mm glass test tubes. Lipids and glycosphingolipids were extracted with 2:1 HPLC-grade chloroform/methanol. The lipid extracts were dried down and purified and fractionated by silica solid phase extraction (Grace Alltech), as described previously (66). The dried protein pellet was digested with trypsin and chymotrypsin (Sigma), as described previously (67). Following heat-inactivation, N-glycans were released with N-glycanase (Prozyme, Hayward, CA) at 37°C for 48 h. Released N-glycans were separated from peptides and O-glycopeptides via C18 solid phase extraction (Sep-Pak, Waters, Milford, MA), as described previously (68). Purified, released N-glycans were reduced via borane/ammonia reduction, as described earlier (69). The reduced N-glycan pool was further purified via graphitized carbon solid phase extraction (Envi-Carb, Supelco, Bellefonte, PA). O-Glycans were released from the dried peptide/O-glycopeptide fraction via reductive-elimination, as described earlier (70). Released O-glycans were purified via cation exchange and porous graphitized carbon solid phase extraction. Intact glycosphingolipid fractions, N-glycan fractions, and O-glycans were permethylated via spin column permethylation (Harvard Apparatus, Holliston, MA), as described earlier (71). Permethylated glycans were purified via liquid-liquid extraction with dichloromethane (EMD), 0.5 M aqueous sodium chloride (Fisher). Extracted permethylated glycoprotein glycans were dried and reconstituted in 50:50 HPLC-grade methanol/water for MS analysis. Intact permethylated glycosphingolipids were dried and reconstituted with 2:1 HPLC-grade methanol/water for MS analysis.

Mass spectrometry
Direct infusion MS experiments were performed using an LTQ (Thermo; ion-trap instrument) equipped with a Nanomate (Advion, Ithaca, NY). Each derived fraction (GSL, O-glycan, and N-glycan) was infused separately and analyzed independently. Normalized collision energy was set to 35% for the LTQ, with activation q set to 0.250 and activation time set to 30 ms. Peak selection for sequential disassembly was made manually; isolation widths were generally set to 2.0 m/z, and isolation was centered at m/z values ϳ0.5-1.0 units higher than the monoisotopic peak to capture the entire isotopic envelope. All ions are sodium adducts, unless otherwise noted. Interpretation of mass spectral data sets was made manually. Where relevant and possible, and when standard materials and mass spectra were available, the unknown substructure spectra were compared with standard spectra, as described previously (45,72). For each spectrum, normalization level and signal average time are indicated. Liquid chromatography-MS analysis of O-glycans was performed using a reversed-phase (C18) column (Thermo BDS Hypersil, 1 ϫ 150 mm) with a 0.1% formic acid/ acetonitrile gradient to elute the reduced and permethylated O-glycans. A Surveyor MS pump and Surveyor autosampler were employed in conjunction with a VelosPro mass spectrometer (all from Thermo). Peak areas of molecular ions were used