Glycan-Protein Interactions Determine Kinetics of N-Glycan Remodeling

A hallmark of N-linked glycosylation in the secretory compartments of eukaryotic cells is the sequential remodeling of an initially uniform oligosaccharide to a site-specific, heterogeneous ensemble of glycostructures on mature proteins. To understand site-specific processing, we used protein disulfide isomerase (PDI), a model protein with five glycosylation sites, for molecular dynamics (MD) simulations and compared the result to a biochemical in vitro analysis with four different glycan processing enzymes. As predicted by an analysis of the accessibility of the N-glycans for their processing enzymes derived from the MD simulations, N-glycans at different glycosylation sites showed different kinetic properties for the processing enzymes. In addition, altering the tertiary structure context of N-glycan substrates affected N-glycan remodeling in a site-specific way. We propose that differential, tertiary structure context dependent N-glycan reactivities lead to different glycan structures in the same protein through kinetically controlled processing pathways.


Introduction
N-glycoprotein biogenesis in eukaryotes is initiated in the Endoplasmic Reticulum (ER) by the oligosaccharyltransferase, an enzyme complex which covalently links a uniquely defined oligosaccharide, Glc 3 Man 9 GlcNAc 2 , to the side-chain amide nitrogen atom of asparagines within -N-X-S/T-sequons. It thereby modifies a large number of proteins, many of them at multiple sites [1][2][3][4]. After the transfer to the protein, processing of the N-linked glycans is initiated by ER-localized hydrolases. The removal of the three glucoses is hereby coupled to the folding of the glycoproteins by providing ligands for lectin chaperones such as calnexin or calreticulin [5,6]. The quality control of protein folding relies on glycans and the information they provide about the conformational state of the covalently bound protein [7].
After the exit from the ER, N-glycoproteins are further processed by Golgi specific hydrolases and transferases that generate the final structures of N-glycans [8]. This remodeling pathway is characterized by individual reactions that rarely go to completion and the processing of a glycan being different for each site of the glycoproteome. Consequently, a site-specific heterogeneity of N-glycan structures is observed. Early on, it was suggested that this differential processing might be due to the tertiary structure of the glycoprotein. In 1984, Savvidou et al. hypothesized that the decreased amount of bisecting N-glycans on one specific glycosylation site of human IgG was due to specific interactions between this glycan and the protein [9]. In the following, several NMR studies demonstrated interactions between N-glycans and the protein surface, which are mostly facilitated by the reducing-end GlcNAc of an N-glycan [10,11]. A most recent example of tertiary structure context dependent glycan processing is the discovery that only a single of the eight N-glycans in the filamentous urinary glycoprotein uromodulin (UMOD) remains a high-mannose type glycan, while all other N-glycans are further processed to complex-type N-glycans. It is however the single high-mannose type UMOD glycan that mediates encapsulation and aggregation of uropathogens by UMOD filaments via interactions with mannoside-specific pilus lectins from the pathogens [12]. In addition, the rise of computational glycobiology allowed the simulation of glycan-protein interactions [13,14] and indicated that these interactions could reduce the accessibility of the glycan to glycan-processing enzymes [15]. Consequently, glycan-protein interactions are considered a major determinant of N-glycan microheterogeneity [15,16]. Several studies have used this knowledge to engineer glycoproteins by site-directed mutagenesis. Chen et al. introduced new glycan-protein interactions into IgG and could thereby significantly improve the stability of IgG against thermal and low pH induced aggregation [17]. In contrast, site-directed amino acid replacements disrupting interactions between the glycan and the protein lead to improved processing of the "freed" N-glycans [18,19].
Understanding site-specific N-glycan processing as it occurs in vivo requires a detailed knowledge of the specificity and localization of the individual hydrolases and glycosyltransferases [20]. After the removal of the three terminal glucoses in the ER, the α-1, 2-mannosidase ER mannosidase I (ER Man I) removes the terminal mannose from the Bbranch of the N-glycan. Even though ER Man I has a high specificity towards this terminal mannose, it is capable of trimming all α-1, 2-linked mannoses from an N-glycan [21,22]. In the Golgi, the glycan is further processed by Golgi Mannosidase I (GM I). Belonging to the same glycoside hydrolase family as ER Man I, this enzyme is also able to remove all α-1, 2linked mannoses from the N-glycan [23]. When confronted with a Man 9 GlcNAc 2 glycan, GM I works least efficiently on the terminal B-branch mannose, the preferred mannose of ER Man The resulting Man 5 GlcNAc 2 glycan is further processed by N-acetylglucosaminyltransferase I (GnT I), which transfers one GlcNAc to the A-branch of the glycan and uses UDP-GlcNAc as a donor substrate [25]. Its action is essential, as the transfer of a GlcNAc initiates the formation of hybrid N-glycans [26]. The generated GlcNAcMan 5 GlcNAc 2 serves as a substrate for Golgi mannosidase II (GM II) which can cleave two mannoses with different glycosidic linkages (α-1, 3 and α-1, 6 linked), of the B-and the C-branch, with a single catalytic site.
We used yeast protein disulfide isomerase (PDI) as a model protein with five N-glycosylation sites to investigate site-specific N-glycan processing in the context of an intact glycoprotein [15]. We performed in-depth molecular dynamics (MD) simulations to analyze the dynamics and the interactions of the N-linked glycans and experimentally addressed site-specific processing by ER Man I, GM I, GnT I and GM II in vitro. Initial velocities and K M values demonstrated that the glycan of each glycosylation site represents a unique substrate to glycan-processing enzymes. MD simulations explained the site-specific properties that were primarily determined by protein-glycan and glycan-glycan interactions. Altering the protein structure changed site-specific glycan processing, validating the conclusion that intramolecular protein/glycan interactions slow or even prevent individual steps of glycan processing. Figure 1A schematically illustrates the chemical composition of a Man 9 GlcNAc 2 glycan, the most abundant structure bound to the glycosylation sites of PDI in the ER. The U-shaped PDI structure shown in figure 1B consists of four thioredoxin-like domains termed a, b, b', and a', of which the domains a and a' possess a catalytic cysteine pair (CGHC in figure 1B) [30].

Molecular dynamics simulations of PDI
Additionally, the a-domain contains a structural disulfide bond (C-C in figure 1B). The glycosylation sites 1 and 2 are located on the a-domain, while the b-domain contains sites 3 and 4. Site 5 is distantly located from the other glycosylation sites on the a'-domain.
We simulated an aggregate sampling of 75 µs for the full-length glycoprotein with

Conformations of PDI glycans
We distinguished four categories in total for the branch accessibility. The two categories with 'free' glycan classification with or without exposed branch(es) were labeled as 'free & exposed branch' and 'free & collapsed branch', respectively. The other two categories differentiate glycan conformations in 'contact' with the protein environment but with or without branch exposure, respectively labeled as 'contact & exposed branch' and 'full contact'. Thus, the classification 'exposed' indicated glycan structures, for which a branch was stretching into the solvent while the others could still be interacting with the protein, neighboring glycans, or other branches within the same glycan. From the enzymatic point of view, extended glycan conformations are preferred because the catalytic sites in GH47 αmannosidases are known to be deep funnels binding only one extended branch at a time [33]. Hence, the 'contact & exposed branch' conformations could be more beneficial for the deep binding funnels than the 'free & collapsed branch' conformations.
The accessibility assessment of sites 1-5 is shown in figure 1C. Sites 1 and 3 exhibited the largest populations of 'free' conformations in which the glycan was barely interacting with its surrounding protein environment or neighboring glycans. Site 2 also showed a clear but reduced fraction of 'free' conformations but had a considerable amount of 'exposed' microstates. Interestingly, a common pattern of 'exposed' branches was shared among sites 1-3 and 5. The 'exposed' microstates were dominated by 'exposed' A-branch conformations, the most flexible branch. A generally minimal exposure of the B branch is related to its central location within the glycan. Naturally, the percentages of exposed conformations declined when considering an increasing number of monosaccharides per branch. This gradient was site-specific. In contrast, the same assessment of the glycan on site 4 showed a significantly different pattern. The percentage of 'full contact' conformations on site 4 was most dominant across all levels of branch lengths, while the fraction of 'free' conformations was the lowest compared to all other sites. Furthermore, the amount of conformations in which a single branch is completely solvent exposed (i.e. 'free & exposed' conformations) was reduced. In contrast, the preference of A and C branch exposure over the B branch was lost, and all three branches had similar fractions of 'exposed' conformations. In summary, our glycan-centric, quantitative analysis of the microstates of the MSMs suggested that the Man 9 GlcNAc 2 was least accessible on site 4, had a tendency to branch exposure but contacting conformations on sites 2 and 5, while sites 1 and 3 exhibited patterns with large solvent exposure. These observations indicated site-specific differences between the five (Man 9 GlcNAc 2 ) sites in their reactivity with ER Man I despite their identical chemical structures.
The above analysis of individual glycan conformations was not accounting for the particular contacts and interactions with the glycans' environments that would lead to the 'free', 'contact', and 'exposed' classifications. For instance, as illustrated in figure 1B, sites 1 and 3 are in close proximity to each other such that glycan-glycan interactions contributed dominantly to the fraction of 'contact' conformations. Hence, a competition of site 1 and 3 during glycan-enzyme interaction could reduce the trimming of the branches on either site. Figure 1B shows further that the glycan on site 5 can easily extend to free conformations but potentially forms frequent interactions with the acidic, C-terminal α-helical PDI segment. 8 Thus, while the amount of free conformations was clearly affected by the glycan-protein interactions at site 5, the individual branches are still exposed during the contact with the neighboring α-helix. In addition, the convex and concave surrounding protein surface topology at site 2 and 4, respectively, pose different glycan-protein contact possibilities ( figure 1B and figure EV1). At site 2, the surrounding protein surface has a positive curvature such that the branches are easily extended when the glycan is 'free' or in 'contact' ( figure   EV1). In contrast, the concave protein surface around site 4 hinders 'exposed' branches in the 'contact' and 'free' conformations.

Experimental setup
To experimentally test the predictions offered by the MD simulations, we turned to an in vitro biochemical analysis of N-glycoprotein processing. We therefore analyzed the kinetics of N-glycan maturation processes of the ER and Golgi (figure EV2A) by incubating PDI with the N-glycan processing enzymes ER Man I, GM I, GnT I, and GM II. PDI was prepared for (mainly) uniform Man 9 GlcNAc 2 , Man 5 GlcNAc 2 , or GlcNAcMan 5 GlcNAc 2 glycosylation corresponding to the respective enzyme's glycan specificity (figure EV3).
After incubation of glycosylated PDI with the respective remodeling enzyme, the site-specific glycoform distributions were obtained by mass spectrometry (MS) of tryptic glycopeptides (figure EV2B). Hence, we observed that for all four enzymes, despite being hydrolases or transferases, site 4 was processed slowest.

Michaelis-Menten analysis of processing kinetics
To quantify the site-specific differences in glycan processing kinetics for the one-step We observed initial velocities increased with PDI concentrations towards saturation (table EV4-5, figure EV4). While the initial velocities of GnT I approached v max for all sites, they still increased nearly linearly on site 4 for ER Man I. Consequently, determination of catalytic parameters of ER Man I was not possible for site 4. The data, however, implied that K M of ER Man I for site 4 is likely at least one order of magnitude higher than for all the other sites.
Also for GnT I, site 4 showed the highest K M value (table 1).
In addition to K M , k cat values were calculated for ER Man I and GnT I, showing the highest turnover of substrate molecules on site 1 and 2, respectively. For both enzymes, k cat /K M was highest for site 2, indicating that the glycan from site 2 was the preferred substrate of ER Man I and GnT I.
In order to compare initial velocities between sites, we normalized them by the initial velocity of site 2. These relative initial velocities were averaged over three PDI concentrations and reveal that the trimming of the oligosaccharide by ER Man I was approximately 10 times slower at site 4 compared to site 2 (figure 3C). For GnT I, site 4 reached only about 7% of the initial velocity of site 2 (figure 3D).

Rate constants describing appearance and disappearance of intermediate structures
Since GM I and GM II catalyze multiple, consecutive trimming reactions, each processing step was studied separately by recording the disappearance of the substrate, the transient accumulation of reaction intermediates and the formation of the final product. For fitting the dependence of the concentrations of all species on reaction time, we assigned apparent, first-order rate constants (k 1 -k 4 ) to each reaction step as an approximation ( figure 4A and B). The global fits of Man 9 GlcNAc 2 processing at sites 1-3 and 5 by GM I agreed reasonably well with the experimental data (figure 4A). The largest deviations from the fit were observed for the kinetics of formation of the final product Man 5 GlcNAc 2 . This indicated that the processing mechanism might be more complex than a consecutive 4-step mechanism and might include branch points and parallel pathways [24].
We therefore extended the reaction mechanism by adding a branch point after Man 7 GlcNAc 2 , assuming the formation of two different Man 6 GlcNAc 2 isomers (appendix figure S1, appendix table S1). The corresponding six-parameter fit indeed agreed better with the experimental data, but we consider the results underdetermined, because we could not experimentally distinguish the different Man 6 GlcNAc 2 isomers.
As shown above for GM II (figure 2D), hardly any conversion of GlcNAcMan 5 GlcNAc 2 to GlcNAcMan 3 GlcNAc 2 could be observed on site 4. To determine the affected mannose trimming step, we measured kinetics of GM II (67 nM) for the processing of the substrate GlcNAcMan 5 GlcNAc 2 -PDI (20 μM). The data of each glycosylation site was fitted according to a consecutive mechanism with two apparent, first-order rate constants k 1

Influence of the protein structure on site-specific mannose trimming
Our atomistic MD simulations revealed an essential role of the tertiary structure context for the site-specific substrate properties of identically composed glycans. PDI contains two catalytic disulfide bonds and one additional, structural disulfide bond (figure 1B) [30, 34].
Nearly complete formation of these disulfide bonds (96%) was confirmed by Ellman assay.
To probe the influence of glycoprotein structure on site-specific processing kinetics, we altered the structure of PDI by reducing and subsequently alkylating the three disulfide bonds. The alkylation resulted in slight conformational changes as detected by far-UV circular dichroism (CD) spectroscopy (figure 5A). As this structural change was by no means comparable to that occurring upon complete denaturation of PDI with 6 M guanidine hydrochloride, the structure of reduced and alkylated PDI still remained native-like.
Reduced and alkylated as well as native, oxidized Man 9 GlcNAc 2 -PDI (20 μM) was then used as a substrate for GM I (0.1 μM) (figure 5B). Site-specific formation of the final product Man 5 GlcNAc 2 was monitored and compared to the respective site on native PDI. We observed that site 4 was processed more efficiently on the reduced and alkylated protein, whereas site 1, 2 and 5 glycans were hydrolyzed slower. For the site 3 glycan, no difference in processing by GM I was observed.

Discussion
We demonstrated that for a given enzyme, glycan processing at the five sites of PDI occurred with different kinetics. For all the enzymes tested, site 4 was processed the slowest, while sites 2 and 5 were processed most efficiently. This order of reactivity was also observed for the same glycoprotein in vivo in insect and CHO cells [15,19]. We therefore concluded that identical glycans at different glycosylation sites presented different substrate properties to the processing enzymes.
We used MD simulations to study the glycan conformations and accessibility on the protein surface in order to explain the site-specific substrate properties of the N-linked glycans. An effective enzyme accessibility of 'free' and/or 'exposed' glycan conformations exhibited a clear site-specificity of the substrate availability and can be qualitatively compared to the enzyme accessibility through the particular structure and glycan-enzyme binding modes. An N-linked glycan on a convex protein surface represents an ideal substrate for a given enzyme, with the affinity solely determined by the glycan-enzyme interaction. Indeed, for ER Man I and GnT I we found the site 2 glycan to be the preferred substrate (highest k cat /K M value). However, the relative initial reaction rates for the other sites differed between the two enzymes, showing that the interpretation of the 'in contact & exposed' glycan conformations in terms of accessibility as well as the influence of the protein surface topology depends on the exact glycoprotein-enzyme complex. For site 4, we noted a higher K M value for both enzymes, indicative for additional protein-glycan interactions due to the concave nature of the protein surface at this glycosylation site. Accordingly, the structural differences of ER Man I and GnT I explain the different site-specific effects on the respective K M values. When the atomistic details of these complexes will hopefully become available in the future, we anticipate that a quantitative and qualitative analysis of glycans based on MD simulations can be further refined.
We perturbed glycan-protein interactions by alkylation of PDI. Similarly, denaturing the glycoprotein soybean agglutinin with 8 M Urea improved the processing of its N-glycans by ER Man I [35]. However, our data on reduced and alkylated PDI suggest that subtle structural differences can already have site-specific effects: well processed and therefore probably easily accessible N-glycans from sites 1, 2 and 5 showed slower processing kinetics upon alkylation. We hypothesize that the change in protein structure featured new interactions, not present in the native PDI, between the glycans from sites 1, 2 and 5 and the PDI surface.
For the site 4 glycan, on the other hand the slight change in protein structure improved processing significantly, while there was no effect detectable for site 3.
Within the framework of our hypothesis, folding intermediates in the ER or conformational isomers of folded proteins in the ER and Golgi may represent distinct substrates for sitespecific glycan processing in kinetically controlled processing pathways. The differential processability of a defined N-linked glycan might even display the folding status of the covalently linked protein in processes such as the quality control pathway of protein folding in the ER [5].
Our detailed biochemical analysis allowed us to follow enzymes that perform multiple processing steps like GM I and GM II. We identified the rate-limiting step in trimming of a Man 9 GlcNAc 2 to Man 5 GlcNAc 2 by GM I to be the last step (Man 6 GlcNAc 2 to Man 5 GlcNAc 2 ).
Lal et al. argued that the last mannose trimmed by GM I is the terminal mannose from the B- 15 branch of the glycan, which is in vivo trimmed in the ER by ER Man I [24]. In case of PDI, this last trimming step was greatly impaired on site 4.
Trimming of two mannoses from GlcNAcMan 5 GlcNAc 2 by GM II is an essential step in the conversion of hybrid to complex N-glycans [36]. Our analysis showed that GM II has, compared to the other enzymes tested, the lowest activity on site 4 with a decrease in activity of two orders of magnitude relative to the other sites, which explained why a secreted version of PDI produced in CHO cells showed the highest percentage of hybrid glycan structures on site 4 [19]. However, our analysis showed that GM II was able to process the site 4 glycan to some degree, indicating a time and/or enzyme-limited process in vivo.
The eukaryotic secretory pathway is organized such that glycoproteins are exposed for a limited time to processing enzymes located in different compartments of the pathway [37-39]. Therefore, site-specific initial processing velocities are a determining factor for N-glycan processing in vivo, together with the residence times of glycoproteins in the individual compartments of the secretory pathway. In such a kinetically controlled system, small alterations of enzyme levels, as observed for example during cellular differentiation in a multicellular organism, can have strong qualitative and quantitative effects on the glycoproteome. Site-specific glycan structures will be affected differently on proteins with multiple glycosylation sites. Therefore, a quantitative glycoproteomics analysis will become a necessity in order to understand the functional properties of glycoproteins.

MD simulations
The

Markov state modeling
Inverse distances of glycan residues to vicinal hexoses and amino acids (with hydrogen bond donating/accepting side chains) were chosen as features for the MSMs, as described before [13,14]. Each glycan was individually analyzed using a site-specific choice of considered neighboring glycans and protein surface residues for the (inverse) distance calculations (table EV1). For the distances involving hexoses, the respective O5 atom positions were taken. For amino acids, a side-chain specific atom position was considered (listed in table EV2). Note that in contrast to the previous MSMs of glycoproteins [13,14], where very few features were manually selected for the sake of modeling ease, our featurization includes all spatially reachable amino acid and carbohydrate moieties.
In the next step, a dimensionality reduction was performed using the principal component analysis (PCA) and choosing the first ten dimensions (n = 10). Subsequently, the hierarchical volume-scaled common nearest neighbor (vs-CNN) clustering algorithm was used to discretize the trajectory into conformational microstates, i.e. clusters [55]. We started by finding an initially large cutoff R 0 , such that 99% of the data was clustered, while the similarity N = 10 was fixed. For sites 1-5 different values were respectively obtained for To quantify the accessibility or exposure of a given monosaccharide in a glycan branch, the exposure score x (x-score) was defined as the combination of the individual measures as follows, where ⟨ solv ⟩ is the average SASA value, ⟨ ℎ ⟩ the average number of hydrogen bonds, and ⟨ ⟩ the average number of contacts. The x-score was calculated for each glycan branch separately, considering the varying length of the branches.
We classified 'free' and 'contact' microstates via the number of atomic contacts across all full-length glycan branches, i.e., tot = ⟨ ⟩ + ⟨ ⟩ + ⟨ ⟩ . For a given threshold N c thresh < N c tot the glycan conformations within the microstate were considered in 'contact'.
Hence, for N c thresh ≥ N c tot the glycan conformations of the microstate were considered to be 'free'. Additionally, branch with x-score s x Y in a given microstate was considered 'exposed' if for a given threshold s x thresh < s x Y . In combination with the MSMs, the relative abundance of particularly classified conformations was calculated as the ensemble average based on the stationary probability distribution from the MSMs. For example, the fraction of exposed conformations is P exposed = ∑ P i i Θ(s x Y,i − s x thresh ) where Θ(x) is the Heaviside step function.
Also, combinations of 'free' and 'contact' with the 'exposed' classifications, respectively, were considered, as P contact&exposed = ∑ P i i Θ(s

Constructs
From the PDI expression construct pRG85 [15], the C-terminal sequence "LELQLEHDEL" was removed by using the primers prHI_14_fw and prHI_15_rev (  Figure appendix S2 shows that freezing and thawing of PDI had no influence on the site-specific processing of its N-glycans.

Production and purification of glycan processing enzymes
High-Five TM cells were infected with recombinant baculovirus stocks for either ER Man I, GM I or GnT I like described before for PDI. After 72 hours, the cell culture was centrifuged at 3500 rcf for ten minutes, supernatants were filtered using 0.

Enzyme in vitro assays
All in vitro assays were performed with PDI as a substrate and either ER Man I, GM I, GnT I or GM II in an enzyme specific activity buffer (table appendix S3). Immediately after addition of the enzyme to PDI, the reaction mix was incubated at 37°C and shaken at 500 rpm.
For in vitro assays with ER Man I and GM I, PDI purified from High-Five TM cells treated with kifunensine was used as a substrate. PDI therefore showed mainly Man 9 GlcNAc 2 on all sites, presenting a homogenous glycosubstrate for the tested enzyme ( figure EV3A). At indicated time points, samples containing a minimum of 50 µg of PDI were taken. To stop the reaction each aliquot was mixed with trichloroacetic acid (15% final concentration) and kept on ice for five minutes. PDI was subsequently pelleted, washed and stored as described by Hang et al. [15].

Sample preparation and MS measurement
Protein pellets were dissolved in 50 µl of 8 M Urea and loaded onto a 30 K centrifugal filter unit (Merck). Preparation for MS was done as described in [64]. acetonitrile with 0.1% formic acid and analyzed by one of the two methods described below: Either by a calibrated LTQ-Orbitrap Velos mass spectrometer (Thermo Fischer) coupled to an Eksigent-Nano-HPLC system (Eksigent Technologies) like described in [15].

Glycoform quantification
Spectra obtained were analyzed with XCalibur 2.2 sp1.48 (Thermo Fisher) like described previously [15]. For quantification, extracted ion chromatography of all glycoforms were plotted by their unique mass over charge (m/z) ratio, based on a previous study by Losfeld et al. (table appendix S4) [19]. Peak area was defined manually and integrated by the program.
The relative amount of each glycoform located on the same peptide backbone was calculated as shown before by Hang et al [15].

Kinetics of site-specific glycan processing by GM I and GM II
For GM I, the glycan processing kinetics of each site were globally fitted using Dynafit [66] according to the following consecutive mechanism where the apparent rate constants k 1 , k 2 , k 3 The apparent rate constants k 1 and k 2 were shared among the three datasets of each glycosylation site.

Reduction and alkylation of PDI
Reduction and alkylation of PDI was done as described earlier for preparation of proteins for MS [64]. The alkylation of cysteines was highly efficient as only alkylated peptides were detected by MS.

Circular dichroism
To determine if reduction and alkylation of the protein changed the protein structure, far-UV circular dichroism (CD) spectra were recorded. PDI was purified from kifunensine treated         The convex and concave protein surface at sites 2 and 4 represent different glycan accessibility.

A:
The positive curvature of the protein surface at site 2 allows extended branch conformations if the whole glycan is classified as 'free'.

B:
In contrast, the glycan at site 4 is confined by the concave protein surface environment, such that even 'free' conformations are still dominated by collapsed conformations. Hence, at site 4 the steric hindrance of a concave protein surface diminishes the accessible space for 'free & exposed' glycan conformations C: Extended branch conformations are particularly 'exposed' at site 2 even if the glycan is in 'contact'.
D: While at site 4 individually 'exposed' branches are still relatively buried inside the concave protein environment if the glycan is in 'contact' with the protein surface.