The importance of N-glycosylation on β3 integrin ligand binding and conformational regulation

N-glycosylations can regulate the adhesive function of integrins. Great variations in both the number and distribution of N-glycosylation sites are found in the 18 α and 8 β integrin subunits. Crystal structures of αIIbβ3 and αVβ3 have resolved the precise structural location of each N-glycan site, but the structural consequences of individual N-glycan site on integrin activation remain unclear. By site-directed mutagenesis and structure-guided analyses, we dissected the function of individual N-glycan sites in β3 integrin activation. We found that the N-glycan site, β3-N320 at the headpiece and leg domain interface positively regulates αIIbβ3 but not αVβ3 activation. The β3-N559 N-glycan at the β3-I-EGF3 and αIIb-calf-1 domain interface, and the β3-N654 N-glycan at the β3-β-tail and αIIb-calf-2 domain interface positively regulate the activation of both αIIbβ3 and αVβ3 integrins. In contrast, removal of the β3-N371 N-glycan near the β3 hybrid and I-EGF3 interface, or the β3-N452 N-glycan at the I-EGF1 domain rendered β3 integrin more active than the wild type. We identified one unique N-glycan at the βI domain of β1 subunit that negatively regulates α5β1 activation. Our study suggests that the bulky N-glycans influence the large-scale conformational rearrangement by potentially stabilizing or destabilizing the domain interfaces of integrin.


Results
Distribution of the N-linked glycans on integrins. The potential N-linked glycosylation sites of integrins were predicted based on the presence of the consensus NXT/S sequons (X is any amino acids except proline). As shown in Fig. 1, the putative N-linked glycosylation sites are distributed among almost all the extracellular subdomains of both α and β subunits within the headpiece and leg domains (Fig. 1). Of the 18 human α integrins, half of them contain an extra ligand-binding αI (inserted) domain ( Fig. 1A-C). The numbers of N-linked glycosylation sites range from 5 (α IIb and α 7 ) to 16 (α 8 ) among the αI-less α-subunits (Fig. 1B), and from 10 (α 2 , α D and α X ) to 26 (α 1 ) among the αI-containing α-subunits (Fig. 1C). The β-subunits have relatively less N-glycan sites compared with the α-subunits, ranging from 5 (β 4 ) to 12 (β 1 ) sites (Fig. 1D). The locations, as well as the numbers of N-glycan sites, vary substantially even within the same subdomains among both α and β subunits. Interestingly, the N-glycan sites are mostly abundant in the leg domains of both α and β subunits, especially in the calf-1 and calf-2 domains of α-subunits (Fig. 1B,C). The α M and α 1 subunits have the most abundant N-glycans (12 sites) in their calf-1 and -2 domains (Fig.1C). Some of the N-glycan sites, such as the ones adjacent to the transmembrane domains of both α and β subunits are relatively conserved (Fig. 1B-D).
Effect of the loss of individual N-glycan sites of α IIb subunit on α IIb β 3 integrin expression and ligand binding. α IIb β 3 integrin has been used as a prototype in understanding integrin structure and function 32 . The α IIb subunit has 5 predicted N-glycan sites: two in the β-propeller domain, one in the thigh domain, one in the calf-1 domain and one in the calf-2 domain (Fig 2A). Three of them have been visualized in the crystal structure ( Fig. 2A). We removed the individual N-glycan site of α IIb subunit one by one by the glutamine substitution. Each Asn to Gln mutant of α IIb was co-expressed with wild-type β 3 subunit in HEK293FT cells. The α IIb β 3 integrin activation was measured by the binding of ligand-mimetic mAb PAC-1 in the physiological Ca 2+ / Mg 2+ condition or in the external integrin activator, Mn 2+ . Overall, the individual Asn to Gln substitution of α IIb subunit had no significant effect on PAC-1 binding to α IIb β 3 when activated by Mn 2+ (Fig. 2B). The α IIb -N931Q mutation slightly reduced PAC-1 binding (Fig. 2B). Compared with the wild-type α IIb subunit, the cell surface Asn residues are shown as sticks with carbons in cyan. N-glycan residues resolved in the crystal structure are shown as sticks with carbons in magenta. Oxygens and nitrogens are red and blue, respectively. It should be noted that the N-glycan residues visualized in the crystal structure are trimmed ones. The native N-glycan chains could be much longer and more complex. (B,C) Ligand-mimetic mAb PAC-1 binding of the indicated single mutations of α IIb co-expressed with β 3 in HEK293FT cells in the presence of 1 mM Ca 2+ / Mg 2+ (Ca/Mg) or 0.2 mM Ca 2+ plus 2 mM Mn 2+ (Ca/Mn). PAC-1 binding was measured by flow cytometry and presented as mean fluorescence intensity (MFI) normalized to integrin expression (AP3 binding). Data are means ± s.e.m. (n ≥ 3). Twotailed t-tests were used to compare the wild type (WT) and the mutants in the Ca/Mn condition. *P < 0.05; **P < 0.01. expression of α IIb β 3 was decreased up to 20% among the α IIb N15Q, N249Q, N570Q and N680Q mutations. However, the α IIb -N931Q mutation dramatically decreased the cell surface expression to more than 50% (Fig. 2B). We also performed the Asn to Ser mutation, given most of the α IIb N-glycan sites locate at a loop region and the serine residue is potentially modified by O-linked glycans. Interestingly, when co-expressed with wild-type β 3 in HEK293FT cells, the α IIb -N15S and the α IIb -N931S mutations significantly reduced Mn 2+ -induced PAC-1 binding (Fig. 2C). There is an increase of PAC-1 binding with the α IIb -N680S mutation, although it is not statistically significant (Fig. 2C). The α IIb -N249S and α IIb -N570S mutations had no effect on PAC-1 binding (Fig. 2C). The Ser substitutions, especially the α IIb -N931S mutation, had less effect on the α IIb β 3 cell surface expression than the Gln substitutions (Fig. 2B,C). Overall, individual deletion of the α IIb N-glycans had little or no effect on the Mn 2+ -induced ligand binding of α IIb β 3 integrin. Of note, the decreased ligand binding by the α IIb -N15S mutation is due to the Ser substitution not the loss of N-glycan since the α IIb -N15Q and α IIb -N15R mutation had no such obvious effect (Fig. 2B).
Effect of the loss of each individual N-glycan site of β 3 subunit on α IIb β 3 integrin expression and ligand binding. The β 3 subunit has 6 N-linked glycan sites: one in the βI domain, two in the hybrid domain, one in the I-EGF1 domain, one in the I-EGF3 domain and one in the β-tail domain (Fig. 3A), all of which have been resolved in the crystal structure (Fig. 3A). All these N-glycans can move with their attached subdomains during the extension of β 3 integrin (Fig. 3B). Similar to the α IIb Asn to Gln mutations, most of the β 3 Asn to Gln single mutations had little effect on the cell surface expression of α IIb β 3 in HEK293FT cells (Fig. 3C). Remarkably, the β 3 N320Q, N559Q and N654Q mutations, located at the βI, I-EGF3, and β-tail domains, respectively, all significantly reduced the Mn 2+ -induced PAC-1 binding (Fig. 3A-C). In contrast, both the N371Q and N452Q mutations, located at the hybrid and I-EGF1 domains, respectively, significantly enhanced the Mn 2+ -induced PAC-1 binding ( Fig. 3A-C). The β 3 -N320Q mutation in the βI domain had the most dramatic negative effect on PAC-1 binding among all the mutations. However, the combined mutation β 3 -N320Q-N559Q did not further decrease PAC-1 binding, but only further reduced the cell surface expression of α IIb β 3 (Fig. 3C). To test whether the mutational effect on the α IIb β 3 PAC-1 binding is specific to the glutamine substitution, we also mutated the Asn to either Ser or Arg. As seen with the β 3 -N320Q mutation, both the β 3 N320R and N320S mutations remarkably reduced the Mn 2+ -induced PAC-1 binding, although the β 3 -N320S mutation also dramatically reduced the cell surface expression of α IIb β 3 (Fig. 3D). In addition, the β 3 N371R and N452S mutations also increased PAC-1 binding, while the β 3 -N654S mutation decreased PAC-1 binding (Fig. 3D). The β 3 N99Q and N99S mutations had no significant effect on Mn 2+ -induced α IIb β 3 PAC-1 binding (Fig. 3C,D).
Effect of the loss of individual N-glycan sites on α IIb β 3 activation from inside the cell. Integrin activation can be triggered from both outside and inside the cell, namely outside-in and inside-out signaling 35 . Having found that the loss of individual N-glycan sites can exert either negative or positive effect on α IIb β 3 activation induced by Mn 2+ from outside the cell, we asked whether it has the similar effect on integrin inside-out activation, in which the signals are initiated from the cytoplasmic domain and transmitted to the ligand-binding site through large-scale conformational changes. The active mutations in the cytoplasmic domains such as the α IIb -R995A and α IIb -F993A or the overexpression of talin-1 head (TH) domain have been used to mimic integrin inside-out activation [36][37][38] . When co-expressed with the active α IIb -R995A mutation, the β 3 N99Q, N320Q, N559Q and N654Q mutations all significantly reduced the constitutive PAC-1 binding to α IIb β 3 (Fig. 5A), while the β 3 N371Q and N452Q mutations significantly enhanced PAC-1 binding (Fig. 5A). The α IIb -F993A mutation rendered α IIb β 3 more active than the α IIb -R995A mutation did (Fig. 5A,B). Likewise, the β 3 N320R and N559Q mutations significantly reduced α IIb -F993A-mediated α IIb β 3 activation (Fig. 5B). The β 3 -N371R mutation did not further enhance the α IIb -F993A-mediated PAC-1 binding probably because the PAC-1 binding already reached the maximal level (Fig. 5B). Interestingly, the β 3 -N559Q mutation in the I-EGF3 domain exerted the most profound defect on α IIb β 3 inside-out activation (Fig. 5A,B).
We next tested the effect of the loss of single N-glycan sites on TH-induced α IIb β 3 activation. We performed this assay in the presence of α IIb -R995A or β 3 -D723A mutation, which has been shown to greatly enhance the TH-induced α IIb β 3 activation 39 . Consistent with the data above, the β 3 N320R and N559Q mutations significantly reduced, while the β 3 N371R mutation slightly increased TH-induced PAC-1 binding (Fig. 5C). As seen above, the β 3 -N559Q had a more remarkable effect than the β 3 -N320R mutation on TH-mediated α IIb β 3 activation (Fig. 5C). These data demonstrate that the N-glycans can exert both negative and positive effect on α IIb β 3 inside-out activation. Consistent with the Mn 2+ -induced PAC-1 binding, the α IIb -N15S mutation reduced TH-mediated α IIb β 3 activation (Fig. 5D). In contrast, the α IIb -N249S mutation increased TH-mediated α IIb β 3 activation (Fig. 5D). However, it should be noted again that the negative effect of α IIb -N15S mutation might not be directly due to the loss of N-glycan.
Effect of N-glycan deletions on α IIb β 3 integrin conformational change. The integrin affinity for ligand binding is tuned by the long-range conformational changes during integrin activation 19 . The ligand binding to the headpiece induces integrin ectodomain extension from the outside-in direction. On the other hand, activators from inside the cell also induce integrin conformational rearrangement, resulting in the affinity increase for the extracellular ligands. Since our data show that some of the N-glycans affect ligand binding of α IIb β 3 integrin, which requires integrin conformational changes, it is tempting to speculate that the N-linked glycans might exert their effect through regulating the conformations of integrin. To test this possibility, we used two conformation-specific mAbs, 319.4 and 370.3, to report the active conformations of β 3 and α IIb , respectively. Eptifibatide (Ept), a high-affinity ligand-mimetic inhibitor that binds both the resting and active α IIb β 3 , was used as a ligand to induce integrin conformational change from outside. Ept induced the binding of both 319.4 and 370.3 mAbs to α IIb β 3 (Fig. 6). The β 3 N320R and N559Q mutations significantly reduced the Ept-induced binding of both mAbs to α IIb β 3 (Fig. 6A,B), while the β 3 -N371R mutation significantly enhanced the Ept-induced . Two-tailed t-tests were used to compare the wild type (WT) and the mutants in the same condition. *P < 0.05; **P < 0.01; ***P < 0.001. mAb 319.4 but not 370.3 binding (Fig. 6A,B). However, there was no obvious effect of the α IIb N15S, N249S and N931S mutations on the Ept-induced mAb binding (Fig. 6C,D), although the N15S and N931S mutations reduced the Mn 2+ -mediated PAC-1 binding (Fig. 2C). These data indicate that individual N-glycans can affect the ligand-induced conformational rearrangement of α IIb β 3 integrin.
To test the effect of N-glycans on the conformational change of α IIb β 3 integrin upon inside-out activation, we used the active cytoplasmic mutations β 3 -K716A 40, 41 and α IIb -R993A. These mutations constitutively induced the binding of mAbs 319.4 and 370.3 to α IIb β 3 (Fig. 7), indicating the conformational changes of integrin from the inside-out direction. The presence of α IIb N15S, N249S and N931S mutations did not affect the β 3 -K716A-mediated binding of anti-β 3 mAb 319.4 (Fig. 7A). However, the α IIb N15S and N931S but not N249S mutations reduced the β 3 -K716A-mediated binding of anti-α IIb mAb 370.3 (Fig. 7B). In contrast, the β 3 N320R and N559Q mutations reduced the α IIb -R993A-mediated binding of both 319.4 (Fig. 7C) and 370.3 (Fig. 7D) mAbs. The β 3 -N371R mutation had no obvious effect on the α IIb -R993A-mediated binding of both mAbs (Fig. 7C,D) probably because the α IIb -R993A mutation already induced the maximum level of α IIb β 3 activation as shown in the PAC-1 binding assay (Fig. 5B). These data demonstrate the importance of N-glycans in integrin conformational rearrangement during inside-out activation.
Effect of N-glycan deletions on the activation of α V β 3 integrin. The β 3 subunit also forms a heterodimer with α V subunit. Changes in the complexity of N-linked glycosylation have been observed in α V β 3 integrins during the metastatic progression of tumor cells 14 . Having established the functional role of the individual N-glycan sites in α IIb β 3 integrin activation, we further studied the effect of N-glycan deletions on the function of α V β 3 integrin. The attachments of N-glycans have been confirmed in the α V β 3 crystal structure for most of the N-glycan sites (Fig. 8A,B). Human fibronectin (Fn) was used as a physiological ligand for α V β 3 . To avoid the effect from the α 5 β 1 integrin, which is the major Fn receptor, we used the HEK293FT cells with both endogenous α 5 and β 1 subunits being knocked out by the CRISPR/Cas9 technology. Consistent with the α IIb β 3 integrin, the β 3 -N559Q and the β 3 -N654Q/S mutation reduced, while the β 3 -N371Q/R mutation increased Mn 2+ -mediated Fn binding to α V β 3 integrin (Fig. 8C). However, the β 3 -N99Q/S, β 3 -N452Q/S, and even the β 3 -N320Q/R mutation had no obvious effect on α V β 3 Fn binding (Fig. 8C). When co-expressed with the activating α V -GAAKR mutation, which mimics α V β 3 inside-out activation 42 , the β 3 -N559Q but not β 3 -N320R mutation remarkably dampened Fn binding (Fig. 8D). This is also in contrast with the α IIb β 3 integrin, in which both β 3 -N559Q and β 3 -N320R mutations greatly reduced the inside-out activation of α IIb β 3 (Fig. 5B). These data indicate that certain individual N-glycans of β 3 subunit may exert different effects on the function of α IIb β 3 and α V β 3 integrins.
Next, we studied the effect of removal of each individual N-glycan site in α V subunit on the activation of α V β 3 integrin. α V subunit has 13 potential N-glycan sites, 10 of which have been confirmed in the crystal structure (Fig. 8A). As shown in Fig. 8E, compared with the wild-type α V β 3 , most of the individual N-glycan deletions by the N to Q substitutions had no obvious effect either on the cell surface expression or the Fn binding of α V β 3 (Fig. 8E). Among all the mutations, only the α V -N458Q and the α V -N943Q moderately increased Mn 2+ -induced Fn binding to α V β 3 (Fig. 8E), and only the α V -N943Q and the α V -N950Q mutations mildly reduced the cell surface expression of α V β 3 (Fig. 8E). We also did the Ser substitutions for each N-glycan site of α V subunit, but no obvious effect was observed on Mn 2+ -induced α V β 3 Fn binding (Data not shown).
When paired with α 5 subunit, the α 5 β 1 heterodimer has 26 potential N-glycan sites. The function of N-glycans in the α 5 β 1 complex formation and cell surface expression or in α 5 β 1 -mediated cell adhesion, migration, and interaction with other cell surface receptors has been studied using mutagenesis approach 8 . By comparing the locations of the putative N-glycan sites among the integrin β subunit (Fig. 1D), we found that the β 1 subunit has an N-glycan site of β 1 -N343 uniquely residing at the β6-α7 loop of βI domain, which has been determined in the α 5 β 1 headpiece crystal structure (Fig. 8F). A hallmark structural change of the βI domain during integrin activation is the downward movement of the β6-α7 loop and the α7-helix 27,30 . Therefore, we hypothesized that the unique N-glycan of β 1 -N343 at the β6-α7 loop might play a role in regulating α 5 β 1 activation. When co-expressed with the wild-type α 5 subunit in the HEK293FT-α 5 β 1 -KO cells, the β 1 -N343R mutation greatly enhanced the Fn binding to α 5 β 1 both in Ca/Mg and Ca/Mn conditions (Fig. 8G). As a control, the β 1 -N249R mutation distal to the ligand-binding site (Fig. 8F) had no effect on α 5 β 1 Fn binding (Fig. 8G). Thus, the loss of the N-glycan at the β 1 β6-α7 loop facilitates α 5 β 1 integrin activation, indicating a negative regulation by this unique N-glycan of β 1 subunit.

Discussion
The great variation in the number and distribution of N-linked glycosylation sites among integrin subunits add another level of heterogeneity to the very complicated integrin family (Fig. 1). An increasing body of evidence indicates that integrin N-glycans contribute to cell adhesion and migration probably by affecting integrin expression, internalization, and association with other cell surface molecules 5,15,16 . However, a direct connection between integrin N-glycans and activation-dependent ligand binding has been missing. In addition, previous work studied the functional effect of either the overall changes in integrin glycosylation or the combined N-glycan sites such as within the same integrin subdomains 20,43,44 , but the function of each individual N-glycan site has not been well documented. In this study, using the structurally well-characterized α IIb β 3 , α V β 3, and α 5 β 1 as model integrins, we found that the loss of certain individual N-glycan site either reduced or enhanced integrin activation reported by the changes in the binding of ligands or active conformation-specific mAbs, indicating that the N-linked glycosylation can exert both positive and negative effects on integrin function. Among the N-glycan mutations of the α IIb subunit, only the α IIb -N15S mutation largely reduced the Mn 2+and TH-induced α IIb β 3 activation. It seems that the negative effect on integrin activation is not due to the loss of N-glycan of α IIb -N15 since the α IIb -N15R and α IIb -N15Q mutations didn't have much effect on α IIb β 3 ligand binding. A previous study showed that the α IIb -N15Q mutation inhibited pro-α IIb maturation, complex formation, and degradation 45 , but we didn't see an obvious effect on the cell surface expression of α IIb β 3 with this mutation. The α IIb -N15 resides at the blade 7 of β-propeller domain close to the interface formed by the α IIb β-propeller and the β 3 βI domain. It is not readily known why the α IIb -N15S mutation affects α IIb β 3 ligand binding. A similar N-glycan site of α V integrin, α V -N44, locates at the blade 1 of α V β-propeller domain, but its mutation to Gln or Ser had no effect on Mn 2+ -induced α V β 3 Fn binding. Thus, the negative effect of α IIb -N15S on α IIb β 3 ligand binding should be specific to the Ser residue, probably due to the gain of O-linked glycosylation as indicated by the change of molecular weight of α IIb -subunit (data not shown), which may affect ligand binding directly or indirectly through affecting α IIb β 3 conformational change. The N-glycans and O-glycans differ in the composition and the formation of the branches within the glycan structures, which determine the interaction with other molecules 12 . Moreover, the N-glycans also vary significantly in length and complexity 12 . Although the complete deletion of α IIb -N15 N-glycan had no effect on α IIb β 3 ligand binding, the negative effect of α IIb -N15S mutation by the potential gain-of-O-glycan suggests that the structure variations of α IIb -N15 N-glycan may regulate the α IIb β 3 function, which is clearly worth further investigation.
Remarkably, almost all the β 3 single N-glycan deletion mutations affected the α IIb β 3 ligand binding induced either by Mn 2+ or by the activating mutations that mimic integrin inside-out activation. Our structural analysis revealed an interesting pattern of the N-glycan location and its impact on integrin ligand binding. All the three β 3 N-glycans, including N320, N559, and N654, which positively regulate α IIb β 3 activation, locate at the α IIb -β 3 inter-chain interfaces. The β 3 -N320 glycan locates near the interface of α IIb β-propeller and β 3 βI domain and inserts into the interface between the headpiece and the leg domains in the bent conformation of α IIb β 3 (Fig. 3A). Similarly, the β 3 -N559 glycan lodges in the interface of α IIb calf-1/2 and β 3 I-EGF3 domains. The β 3 -N654 glycan resides at the interface of α IIb calf-2 and β 3 β-tail domains (Fig. 3A). All of these interfaces are disrupted during integrin activation due to the headpiece extension and leg domain separation (Fig. 3B) 19 . The hydrophilic and bulky glycan groups may destabilize these interfaces at the bent conformation by repulsive interactions and therefore facilitate integrin conformational change. Consequently, deletion of these wedge-like glycans dampens α IIb β 3 ligand binding and conformational change potentially due to the stabilization of the bent inactive conformation of integrin. This is consistent with the previous studies showing that introducing an artificial N-glycan site into the βI and hybrid domain interface of β 3 subunit 46 or into the thigh and calf-1 domain interface of α IIb subunit 47 renders α IIb β 3 constitutively active by enforcing integrin extension. In contrast, the β 3 N-glycans, attaching to N371 and N452, which negatively regulate α IIb β 3 activation, are located close to the intra-chain interfaces of β 3 subunit. Unlike the β 3 -N559 glycan that inserts into the domain interface, the β 3 -N371 glycan lies on the interface formed by the β 3 hybrid and I-EGF3/4 domains (Fig. 3A), which is disrupted upon the β 3 extension (Fig. 3B). The loss of β 3 -N371 glycan increased α IIb β 3 activation, indicating that this glycan contributes to stabilizing the interface highly possibly by acting like a door bolt. In addition, the loss of the β 3 -N99 N-glycan, which is adjacent to the N-glycan of β 3 -N371 but farther away from the hybrid/I-EGF interface, slightly decreased α IIb β 3 activation. The close contacts between β 3 -N371 and β 3 -N99 N-glycans may influence the conformation of the β 3 -N371 N-glycan, which in turn affects the hybrid/I-EGF interface. Upon deletion of the β 3 -N99 N-glycan, the conformation of the β 3 -N371 N-glycan may further stabilize the hybrid/I-EGF interface at the bent conformation. The β 3 -N452 glycan resides at the opposite side of the acute angle formed by the I-EGF1 and I-EGF2 domains (Fig. 3A). This angle opens to almost 180° after integrin extension (Fig. 3B), which places the β 3 -N452 glycan at the newly formed I-EGF1/2 domain interface (Fig.3B). As a result, the bulky N-glycan of β 3 -N452 may exert repulsive tension to the interface and thus destabilize the extended conformation. Indeed, the removal of β 3 -N452 N-glycan facilitated α IIb β 3 activation. Thus, our data demonstrate a location-specific contribution of individual N-glycan sites in regulating integrin activation largely through affecting integrin conformational changes.
A direct link between integrin N-glycans and the conformational regulation has been missing until a recent study, in which the effect of N-glycans on the conformational equilibria of α 5 β 1 integrin was elegantly investigated 48 . Our data are consistent with their results showing that the complex N-glycans stabilize the extended active conformation relative to the bent resting conformation. It was also proposed that the bulky N-glycans might exert their regulatory roles by crowding or repulsive interactions within the domain interfaces 48 , which is in agreement with the structure interpretations of our data. However, only the combined effects of the overall changes of N-glycans on integrin affinity were examined in the α 5 β 1 study and the results might be a mixture of both negative and positive effects, although the positive effect was obviously dominated 48 . Our data show that individual N-glycan sites exert different effects on the β 3 integrin activation, arguing the importance of investigating the functional role of individual N-glycans in integrin activation.
Although α V β 3 integrin shares the same β 3 subunit with α IIb β 3 , it is influenced differently by the loss of certain β 3 N-glycan sites. Particularly, the β 3 -N320 mutation greatly reduced α IIb β 3 activation but had no effect on α V β 3 activation. Moreover, the α V -N844 glycan, although buried at the interface between the β 3 βI and β-tail domain in the bent conformation of α V β 3 (Fig. 8A), also had no effect on Mn 2+ -mediated α V β 3 Fn binding, consistent with a recent study 29 . The requirement of large-scale conformational changes for α V β 3 ligand binding remains controversial 29,49 , but an increasing body of data supports the importance of integrin extension and headpiece opening in α V β 3 activation 29 . However, since the function of α IIb β 3 is to mediate platelet aggregation for hemostasis, it requires the activity of α IIb β 3 to be strictly regulated 31 , while the fundamental function of α V β 3 is to mediate cell adhesion and migration that are essential for cell survival and proliferation 34 , and thus the activation of α V β 3 might not be as strictly regulated as α IIb β 3 integrin. The different effect of β 3 N-glycans on α V β 3 and α IIb β 3 activation might be attributed to the differences in their biological function. Indeed, the differences in the affinity regulation between α V β 3 and α IIb β 3 had been reported previously 50,51 .
Scientific RepoRts | 7: 4656 | DOI:10.1038/s41598-017-04844-w The natural variants that cause the loss-or gain-of-function due to individual glycosylation mutations have not been reported for β 3 integrins. Nevertheless, a recent study showed that aberrant glycosylation of both α V and β 3 subunits were observed between primary and metastatic melanoma cells, which modify the integrin-mediated tumor cell migration 14 . However, it is not known which N-glycan site is changed during the tumor cell progression. Moreover, N-glycans can have different types depending on their contents, including the high-mannose, hybrid, and complex N-glycans. It has been shown that the composition of N-glycans such as branching and sialylation regulates the function of N-glycans 2 . Considering the importance of individual N-glycan sites in integrin activation as determined in the present study, it will be of great interest to determine if the heterogeneity of certain individual N-glycans affects their regulatory roles in integrin function. Of the 13 N-glycan sites of α V subunit, deletion of each individual site had little effect on α V β 3 expression and activation. It will be interesting to know if the simultaneous removal of multiple N-glycan sites will affect integrin function as shown in the study of α 5 β 1 integrin 20,44 .
Among the 24 human integrins, the function of N-glycans of α 5 β 1 has been relatively well characterized. The N-glycan sites that are important for α 5 β 1 heterodimer formation and biological function have been determined in both α 5 and β 1 subunits 20,44 . Recent studies have mapped the important biological function to several individual N-glycan sites. For example, the N-glycan sites at the α 5 β-propeller domain are important for α 5 β 1 -mediated cell adhesion and migration 21,52 ; one of the N-glycans at the α 5 calf-1 domain is important for the complex formation with EGFR and the inhibition of EGFR signaling 53 ; the N-glycan at the β 1 β-tail domain is important for β 1 activation and interaction with other cell membrane proteins including syndecan-4 and EGFR 54 ; the 3 N-glycan sites at the β 1 βI domain are important for α 5 β 1 complex formation and cell spreading 44 . Furthermore, recent kinetic studies of the correlation between α 5 β 1 affinity and conformation suggest that the complex-type N-glycans of α 5 β 1 help stabilize the high-affinity conformation 48 . In the current study, we identified one N-glycan site at β 1 -N343 in the βI domain, which negatively regulates β 1 activation since the removal of this glycan site rendered α 5 β 1 constitutively active. The β 1 -N343 glycan may directly regulate the conformational change of βI domain by restraining the movement of the β6-α7 loop, or directly regulate ligand binding due to its proximity to the ligand binding site as proposed based on the structure modeling studies 55,56 . We expect to identify more location-specific functions of individual N-glycan sites in α 5 β 1 and other integrins. These N-glycan sites may work in concert to balance the biological activity of integrin.
Understanding of integrin conformation and affinity regulation has been greatly assisted by the mutagenesis studies. Critical residues and interactions that are important to integrin activation have been identified based on the loss-or gain-of-function mutations 57 . By site-directed mutagenesis, our study provides evidence that individual N-linked carbohydrate residues, depending on their structural locations, regulate integrin activation at least in part through influencing the conformational rearrangement. Exactly how these N-glycan "hotspots" are modified and regulated and how they contribute to integrin affinity and biological function need to be studied in more detail using both structural and cell biology approaches.
Cell lines. HEK293FT cells (ThermoFisher Scientific) were cultured in DMEM plus 10% FBS at 37 °C with 5% CO 2 . The α 5 and β 1 subunits double-knockout HEK293FT (HEK293FT-α 5 β 1 -KO) cells were generated by the CRISPR/Cas9 gene editing technology using the α 5 and β 1 CRISPR/Cas9 KO plasmids from Santa Cruz Biotechnology. In brief, cells were transfected with the α 5 KO plasmids for 5-7 days. The α 5 -negative cells were selected by single cell sorting after staining with the anti-α 5 mAb VC5. The established α 5 -KO cells were transfected with the β 1 KO plasmids and the β 1 -negative cells were selected by single cell sorting after staining with the anti-β 1 mAb MAR4. Single cell clones with the lowest expression of both α 5 and β 1 subunits were selected for the following experiments.
Conformation-specific antibody binding. Binding of the active conformation-specific anti-β 3 mAb 319.4 and anti-α IIb mAb 370.3 to the HEK293FT transfectants was performed as described 39,57 . In brief, cells were first incubated with 10 μg/ml biotin-labeled 319.4 or 370.3 in the HBSGB buffer containing 1 mM Ca 2+ /Mg 2+ in the absence or presence of 10 μM α IIb β 3 -specific ligand-mimetic inhibitor eptifibatide at 25 °C for 30 mins, and then washed and incubated with 10 μg/ml Alexa fluor 488-labeled AP3 and Alexa fluor 647-labeled streptavidin on ice for 30 min. AP3 positive cells (expressing α IIb β 3 integrin) were acquired for calculating the MFI by flow cytometry. The 319.4 or 370.3 mAb binding was presented as normalized MFI, i.e. streptavidin MFI as a percentage of AP3 MFI. By this calculation, the binding of conformation-specific mAbs was normalized to total integrin expression.
Statistical analysis. The statistical analysis was performed using the GraphPad Prism software. Two-tailed Student's t-test was used to calculate the p values for comparing the two experimental groups, for example, the wild type and the mutant data under the same condition. The assays were repeated independently at least three times for statistical analysis.
Data Availability. The datasets generated during and/or analysed during the current study are available from the corresponding author on reasonable request.