Shotgun glycomics of pig lung identifies natural endogenous receptors for influenza viruses

Significance

Studies using novel “shotgun glycan microarray” technology identify, for the first time to our knowledge, the endogenous receptors for influenza viruses from a natural host, the pig. Libraries of total N-glycans from pig lung were probed for binding properties using a panel of influenza viruses isolated from humans, birds, and swine. Natural glycan receptors were identified for all viruses examined, and although some displayed the rather broad α2,3 or α2,6 sialic acid linkage specificity conventionally associated with avian or human viruses, other strains were highly specific, revealing a complexity that has not been demonstrated previously. Because pigs are often implicated as intermediate hosts for pandemic viruses, these results and the approaches described will transform our understanding of influenza host range, transmission, and pathogenicity.

Abstract

Influenza viruses bind to host cell surface glycans containing terminal sialic acids, but as studies on influenza binding become more sophisticated, it is becoming evident that although sialic acid may be necessary, it is not sufficient for productive binding. To better define endogenous glycans that serve as viral receptors, we have explored glycan recognition in the pig lung, because influenza is broadly disseminated in swine, and swine have been postulated as an intermediary host for the emergence of pandemic strains. For these studies, we used the technology of “shotgun glycomics” to identify natural receptor glycans. The total released N- and O-glycans from pig lung glycoproteins and glycolipid-derived glycans were fluorescently tagged and separated by multidimensional HPLC, and individual glycans were covalently printed to generate pig lung shotgun glycan microarrays. All viruses tested interacted with one or more sialylated N-glycans but not O-glycans or glycolipid-derived glycans, and each virus demonstrated novel and unexpected differences in endogenous N-glycan recognition. The results illustrate the repertoire of specific, endogenous N-glycans of pig lung glycoproteins for virus recognition and offer a new direction for studying endogenous glycan functions in viral pathogenesis.

Influenza A viruses infect the human population on a yearly basis, causing hundreds of thousands of excess deaths, enormous stress on health care systems worldwide, and extensive economic burden. At unpredictable intervals, viruses with antigenically novel glycoproteins emerge to cause pandemics of even greater concern, and the pig is often implicated as an intermediate host when this occurs (1). The waterfowl that act as natural reservoirs for influenza A viruses harbor 16 antigenically distinct subtypes of the hemagglutinin (HA) surface antigen, and 9 subtypes of the neuraminidase (NA) envelope protein, but the only subtypes to emerge and circulate broadly in humans over the past century are H1N1 in 1918, H2N2 in 1957, H3N2 in 1968, and H1N1 again in 2009. However, the possible emergence of a novel subtype in humans remains a concern, because highly pathogenic H5 and H7 subtype “bird flu” viruses continue to circulate in avian species, and over the past 15 y have infected humans, with limited ability to transmit between humans. The reasons for such host–species restriction and mechanisms underlying cross species transmission are thought to be multifactorial, but the receptor binding properties of the HA surface antigen are of paramount significance.

Since the identification of sialic acids as critical components of influenza receptors (2, 3), a broadly observable phenomenon has been that avian strains generally prefer binding to glycan receptors with sialic acids containing α2,3 glycosidic linkages to the penultimate galactose in the carbohydrate chain, whereas human viruses favor binding to substrates with α2,6 linkages (4–7). The description of mutant viruses for which specific changes in the HA receptor-binding site result in a switch from one binding preference to the other provides clues on potential mechanisms for altered host range (8–14) but perhaps leads to the simplified perception that influenza binds rather indiscriminately, as long as the minimal sialic acid linkage is appropriate. Interestingly, proton-NMR binding data for avian-like or human-like HAs to monosialosides of preferred, or nonpreferred linkage, show binding affinity to be generally very weak, with dissociation constants in the millimolar range and only a two- to threefold difference between cognate pairs and reciprocal pairs (15, 16). This suggests that productive virus attachment results from the cumulative effects of multiple HA receptor interactions in which the consequences of very small differences are amplified in a phenotypically distinguishable fashion. However, little more is known with regard to binding affinity for individual glycans of defined structure or of their presence and distribution in tissues encountered during natural infections.

Initial studies using linkage-specific lectins as an indirect probe for the presence and distribution of potential influenza glycan receptors suggested that the intestinal tract of ducks is rich in α2,3-linked sialic acid (α2,3-Sia) and the upper respiratory tract of humans contains an abundance of sialic acid in α2,6 linkage (17). These observations have been extended considerably, and although the literature on studies applying lectin histochemistry data is complicated, a general consensus suggests that human upper airways contain primarily α2,6-Sia, and the lower respiratory tract contains a mixture with considerable representation of both α2,3-Sia and α2,6-Sia receptor types (18–21) The respiratory tract of swine also contain both α2,3-Sia and α2,6-Sia, which appear to be distributed similarly to humans (17, 22, 23). These observations have been used in support of the hypothesis that the pig may provide an intermediate host for the adaptation of avian strains before emergence in humans or for the genesis of reassortant pandemic viruses capable of human infection following coinfection with avian and human strains (24).

Although lectin histochemistry may provide an indicator for the presence of glycans of particular sialic acid linkage, it does not identify actual glycan receptors. The continued development and use of new tools in carbohydrate chemistry and an expanding range of viruses under scrutiny have made clear that not only sialic acid linkage but underlying glycan composition and structure, chain length, and presentation of receptors also play a role in the complex interactions that lead to productive attachment of influenza to host cells. Defined glycan microarrays have been developed and successfully applied to examine the binding specificity of influenza, as well as other viruses and microorganisms (25–32). However, a limitation of most such arrays is that they are based on chemically or enzymatically synthesized glycan structures and represent only a subset of the endogenous complex glycomes of natural hosts. The restrictions imposed by the synthetic nature of such arrays is emphasized by the observation that mutant influenza viruses have been identified which replicate both in cell culture or in the mouse respiratory tract and yet exhibit no detectable binding to the glycans on the Consortium of Functional Glycomics (CFG) array (33). In addition, recent glycan profiles of swine epithelial cells (34), or tissues from human respiratory tissues (21), demonstrate a diversity of glycan structures that are not comprehensively represented on available glycan arrays. Furthermore, the observation that α2,6-linked, as well as α2,3-linked, sialic acids were identified by mass spectrometry (MS) in human lung glycans indicates that alveolar tropism of influenza virus may not be dictated by its binding only to α2,3 glycans (21). The available data highlight the need for more focused arrays that represent the distribution of natural endogenous receptors derived from tissues in which influenza replicates in natural hosts.

Our group recently developed a strategy of shotgun glycomics to successfully identify naturally occurring glycoconjugate ligands for glycan-binding proteins (GBPs) and serum antibodies (35). In this strategy, summarized in Fig. 1, glycans released from naturally occurring glycoconjugates are labeled with a bifunctional fluorescent tag, fractionated and purified as individual glycans, printed as glycan microarrays, and interrogated with biological targets, such as viruses. The glycans that are strongly bound are considered biologically relevant and are structurally characterized. This novel approach is a shortcut toward defining relevant protein–glycan interactions, because it limits the difficult task of structural characterization of an entire glycome and the laborious chemical synthesis of complex and possibly irrelevant glycans. Using this strategy for the study of influenza virus binding is ideal, because natural glycan microarrays can be prepared from biologically relevant cells and tissues and directly screened with different strains of viruses. In this report, we describe the isolation, fractionation, purification, and interrogation of the N-glycome of the pig lung, a potential site of infection for influenza viruses of avian, swine, and human origin and determine the binding profiles for a panel of influenza viruses that include isolates from each of these host species and represent several HA subtypes. Although some of the expected trends of binding to α2,3- or α2,6-linked sialic acid receptors were observed, the binding profiles did not correlate with those of specific lectins, and receptor recognition was quite selective for particular strains. In several examples, modifications to the glycan backbone were influential, and for several viruses preferential recognition of branched chain di- and triantennary structures suggested a role for multivalent interactions. To our knowledge, this is the first identification of binding characteristics of influenza viruses to specific natural N-glycan derived receptor molecules.

Fig. 1.

Principle of shotgun glycan microarrays from pig lung. Total glycans are isolated from lung tissue glycoproteins or glycosphingolipids, fluorescently tagged and separated by HPLC to generate a tagged glycan library (TGL) of purified fractions collected in 96-well plates. Each glycan fraction is then printed in equal amounts (∼1 nmol) on NHS-activated glass slides and interrogated with fluorescently labeled influenza A virus. The glycans corresponding to bound positions on the slide are retrieved from the TGL and structures determined by MS and other methods to identify the candidate ligands for specific viruses.

Results

Pig Lung N-Glycan Library.

Pig lung tissue was obtained from a commercial vendor or from a germ-free experimental animal, and in each case, 15 g of pig lung tissue yielded ∼3.1 µmol of total N-glycans. Following separation by HPLC, Lampire lung-derived glycans were determined to be a mixture of neutral (∼31%), monosialylated (∼23%), disialylated (∼40%), and trisialylated (∼6%) species (SI Appendix, Fig. S1). To estimate recovery of the total N-glycans released by N-glycanase from the starting material of tryptic glycopeptides, we followed the distribution of mannose, which is a common constituent of N-glycans. N-glycanase digestion released ∼90% of the total mannose, whereas ∼5% remained associated with residual N-glycanase–resistant glycopeptides. We analyzed total permethylated N-glycans in preparations from the two sources of lung tissue using MALDI-TOF (36, 37). The results shown in SI Appendix, Fig. S2A revealed that the N-glycans from the purchased pig lung have compositions that indicate they are dominated by sialylated and fucosylated biantennary complex-type structures (m/z 2,966.9, 2,605.3, etc.). A remarkable signal of predicted triantennary complex-type structures (m/z 3,777.2, etc.) is also observed. Signals of high mannose-type N-glycans, ranging from Man₅GlcNAc₂ to Man₉GlcNAc₂, were also detected in minor abundance (m/z 1,579.8, 1,783.9, 1,988.0, 2,192.1, and 2,396.2). N-glycolylneuraminic acid (Neu5Gc) (m/z 2,635.3, 2,996.7, etc.) and the predicted αGal epitope (Gal-Gal) (m/z 2,809.5, etc.) are also found in the N-glycans. Interestingly, we did not detect significant amounts of glycans with poly-N-acetyllactosamine (LacNAc) repeating structures (-3Galβ1,4GlcNAcβ1-_n). Repeating this analysis on the permethylated N-glycans from the pathogen-free pig lung gave a similar pattern (SI Appendix, Fig. S2B), which indicated the reproducibility of our preparative techniques and that the N-glycome was independent of pathogen exposure.

Pig Lung N-Glycan Shotgun Glycan Microarrays.

The released N-glycans were derivatized with 2-amino-N-(2-aminoethyl)-benzamide (AEAB) as described previously (38) and separated in the first dimension using normal-phase HPLC. Each peak from the normal phase column was rechromatographed by HPLC on a porous graphitized carbon (PGC) column to obtain a pig lung tagged glycan library (PL-TGL) comprised of 96 N-glycan fractions (SI Appendix, Table S1). This library, along with four standards, including glycans with α2,3- and α2,6-linked sialic acid, was printed on N-hydroxysuccinamide (NHS)-derivatized glass slides to produce v1.0 of the pig lung shotgun glycan microarrays (PL-SGMs). The N-glycans were quantified based on fluorescence to estimate relative abundance (SI Appendix, Table S1) and prepare them at 100 µM for printing 0.33 nL of each in replicates of four. Before printing, each entry of the PL-TGL was analyzed by MALDI-TOF to identify the molecular ions in each glycan target. In many cases, the unique mass provided monosaccharide composition, and most fractions contained only one or two glycans, thus indicating the relative purity of the isolated glycans.

Lectin Binding to Pig Lung N-Glycan Microarray.

To confirm that the N-glycans were printed on the PL-SGM, we interrogated the array using biotinylated lectins with defined glycan binding specificity and the results are shown in Dataset S1, which provides the average relative fluorescence units (RFUs), coefficient of variation (%CV), and ranking of the glycans in each analysis, as displayed in the histograms in Fig. 2. The rank of individual glycans in a binding assay was calculated as a percentile of RFU of the glycan divided by the highest RFU values in that assay. Sambucus nigra agglutinin (SNA) binding to the PL-SGM (Fig. 2A) indicates the glycans that possess terminal α2,6-linked sialic acid to galactose (26, 39, 40). Maackia amurensis agglutinin-I (MAL-I) binding, which indicates the presence of terminal sialic acid linked α2,3 to galactose (41), is shown in Fig. 2B. Interestingly, most of the glycans appear to be homogeneous with respect to sialic acid linkage because there was very little overlap in the SNA- and MAL-I–binding profiles. Comparison of the rankings of MAL-I– and SNA-binding glycans (Dataset S1) indicates significant potential overlap in only three glycan fractions, glycans 24, 62, and 63. Binding of these lectins was lost upon treatment of the PL-SGM with neuraminidase, indicating that recognition was sialic acid-dependent and not attributable to sulfate; recent studies indicate that SNA and MAL-I can both interact with 6-O– or 3-O–sulfated terminal galactose residues, respectively (42, 43). The N-glycan PL-SGM was interrogated with other defined biotinylated lectins including Aleuria aurantia lectin (AAL), which binds α-linked fucose in a variety of linkages (44) (Fig. 2C); concanavalin A (ConA), which binds complex-type biantennary, hybrid-type, and high mannose-type N-glycans but not tri- or tetraantennary complex N-glycan structures (45–47) (Fig. 2D); Erythrina crystagalli lectin (ECL), which binds terminal β1,4–linked galactose (48) (Fig. 2E); and Ricinus communis agglutinin-I (RCA-I), which binds terminal β-linked galactose or galactose modified with α2,6-linked sialic acids but not α2,3-linked sialic acids (49) (Fig. 2F). The ECL binding profile after neuraminidase digestion (Fig. 2G) indicates that the terminal sialic acids were predominantly linked to galactose in the glycans presented on the PL-SGM. The behavior of lectins binding to the four control glycans is consistent with the known specificity of the lectins, and, taken together, the data indicate that we were successful in printing N-glycans on the PL-SGM and that these glycans possess both α2,6- and α2,3-linked sialic acid.

Fig. 2.

Lectins binding to pig lung N-glycan microarray. Glycan microarray TGL fractions were incubated with lectins of different specificity to validate the arrays. The lectins used were: SNA (specific for Siaα2,6Gal-R) (A); MAL-I (specific for Siaα2,3Gal-R) (B); AAL (specific for fucose) (C); ConA (specific for biantennary but not triantennary complex N-glycan structures) (D); ECL (specific for terminal galactose) (E); and RCA-I (specific for terminal galactose or α2,6-linked sialic acids but not α2,3-linked sialic acids) (F). The slides were also treated with neuraminidase, followed by binding of ECL (G) to identify glycans with newly exposed galactose sugars to which sialic acids had been linked. Fractions 1–96 are N-glycans isolated from pig lung: 1–14, neutral; 15–47, monosialyl; 48–77, disialyl; and 78–96, trisialyl. Standards include: 97, NA2 (nonsialylated biantennary N-glycan); 98, NA2-Sia2,3; 99, NA2-Sia2,6; and 100, Man5.

Influenza Virus Binding to N-Glycans of the PL-SGM.

To explore the range and diversity of potential influenza virus natural glycan receptors present in the pig lung, a panel of viruses isolated from human, swine, and avian species was examined. The viruses selected represent a broad spectrum of viruses encompassing differing species of origin, time, and geographic location of isolation and various HA subtypes including H1, H2, H3, and H6. (Table 1). Included in the panel were seasonal human isolates of both an H1N1 subtype (A/Pennsylvania/08/2008) and an H3N2 subtype (A/New York/55/2004), as well as 2009 human pandemic (pdm) H1N1 strains A/California/04/2009, A/Mexico/4487/2009, and A/Texas/15/2009, which represent viruses that displaced the previously circulating H1N1 seasonal strains (i.e., A/Pennsylvania/08/2008). These 2009 viruses are direct human isolates (not passaged in eggs) from the H1N1 pandemic, which caused different pathogenic symptoms in humans and have strict α2,6-linked sialic acid specificities (50). The contemporary swine viruses of the panel were isolated from pigs at the time of the 2009 human pandemic and are thought to be representative of viruses with related gene constellations that may have given rise to the human pandemic strains. The three swine strains, A/sw/Minnesota/02719/2009 (H3N2), A/sw/Minnesota/02749/2009 (H1N1), and A/sw/Illinois/02860/2009 (H1N2), have several internal genes in common with one another and with human pandemic strains but different surface glycoprotein combinations. The Ruddy Turnstone viruses are low-pathogenicity avian H6N1 and H1N9 subtypes. A/Brisbane/59/2007 (H1N1) is a vaccine strain that has been passaged extensively in eggs, and A/New York/55/2004 (H3N2) is a seasonal virus, which was passaged one to two times in eggs before transfer to MDCK cells. None of the other human or swine viruses analyzed have been passaged in embryonated chicken eggs, a common procedure for growing field or clinical isolates that is known to select for changes in the binding characteristics, including adaptation to α2,3-sialylated receptors (51). We have shown that A/Brisbane/59/2007 (H1N1) has broad specificity for both α2,3- and α2,6-sialylated receptors (50).

Table 1.

Virus strains tested on the microarrays

Virus	Subtype	Host
Avian isolates
A/RuddyTurnstone/DE/650625/2002	H6N1	Avian
A/RuddyTurnstone/DE/650645/2002	H1N9	Avian
Contemporary swine isolates
A/sw/Minnesota/02749/2009	H1N1	Swine
A/sw/Minnesota/02719/2009	H3N2	Swine
A/sw/Illinois/02860/2009	H1N2	Swine
Vaccine strain
A/Brisbane/59/2007	H1N1	Human
Seasonal
A/Pennsylvania/08/2008	H1N1	Human
A/New York/55/2004	H3N2	Human
Pandemic H1N1
A/California/04/2009	H1N1	Human
A/Mexico/InDRE4487/2009	H1N1	Human
A/Texas/15/2009	H1N1	Human

The 11 viruses were labeled with a fluorescent tag and analyzed on the glycan microarray (Version 5.1) developed by the CFG (www.functionalglycomics.org), which contains 610 immobilized chemoenzymatically generated glycans, representing a variety of defined glycan epitopes or determinants found on N- and O-glycans and glycosphingolipid-derived mammalian-type glycans. The binding patterns of the viruses on the CFG array demonstrated that all of the viruses bound strongest to sialylated N-glycans and common N-glycan epitopes (Fig. 3). Each preparation of labeled virus was divided and examined in parallel on the CFG array and on the N-glycans of the PL-SGM. All of the isolates were prepared in this way to maintain consistency between array experiments. None of the viruses bound to neutral glycans, but many mono, di-, and trisialylated N-glycans were bound by the different virus strains (Fig. 3), and each of the viruses displayed a distinct binding pattern to the PL-SGM.

Fig. 3.

Influenza virus binding to glycan microarrays. Eleven viruses have been tested including isolates from vaccine (A/Brisbane/59/2007), human (A/New York/55/2004, A/Pennsylvania/08/2008, A/California/04/2009, A/Mexico/INDRE4487/2009, and A/Texas/15/2009), avian (the Ruddy Turnstone viruses), and swine (A/sw/Minnesota/02719/2009, A/sw/Minnesota/02749/2009, and A/sw/Illinois/02860/2009). (A) PL-SGM array. Fractions 1–96 are N-glycans isolated from pig lung: 1–14, neutral; 15–47, monosialyl; 48–77, disialyl; and 78–96, trisialyl. Standards include: 97, NA2; 98, NA2-Sia2,3; 99, NA2-Sia2,6; and 100, Man5. (B) CFG array. This microarray contains 610 immobilized glycans, representing a variety of defined glycan epitopes or determinants. Only the epitopes containing terminal sialic acid are represented here. The glycans are grouped by terminal linkage with the α2,3 linkage type highlighted in green, the α2,6 linkage highlighted in blue, and the α2,8 linkage highlighted in purple.

Waterfowl isolates.

The binding patterns of A/Ruddy Turnstone/650625/2002 (H6N1) and A/Ruddy Turnstone/DE/650645/2002 (H1N9) were compared on the CFG glycan array with the help of the Cross Analysis tool in the GlycoPattern website (https://glycopattern.emory.edu). This comparison indicated a clear relationship between these two strains. Both avian strains demonstrated a definite specificity for NeuAcα2,3Gal-terminating glycans. Of the two strains, A/Ruddy Turnstone/DE/650645/2002 (H1N9) demonstrated a much more restricted specificity, binding primarily to linear fucose-containing glycans. Interestingly, all of the glycans bound by strain A/Ruddy Turnstone/DE/650645/2002 (H1N9) were also bound by strain A/Ruddy Turnstone/650625/2002 (H6N1); however, A/Ruddy Turnstone/650625/2002 (H6N1) bound many more glycans with lower ranking (14–40 percentile), many of which were branched glycans. When a similar comparison was made between these two strains on the PL-SGM (Fig. 3 and Dataset S2), no such overlap was observed. The A/Ruddy Turnstone/650625/2002 (H6N1) strain bound 34 glycan fractions ranked between 100 and 20 percentile, whereas the four glycans bound strongest (ranked 73–100) by strain A/Ruddy Turnstone/DE/650645/2002 (H1N9) were among the weaker binding glycans for strain A/Ruddy Turnstone/650625/2002 (H6N1).

Swine isolates.

The three swine isolates A/swMinnesota/02749/2009 (H1N1), A/sw/Minnesota/02719/2009 (H3N2), and A/sw/Illinois/02860/2009 (H1N2) were analyzed on the defined glycan microarray from the CFG and the PL-SGM (Fig. 3 and Dataset S2) in parallel. These three isolates were all generally specific for α2,6-linked sialic acids and bound strongest to sialylated linear and branched lactosamine- or polylactosamine-containing glycans on the CFG array. When analyzed on the PL-SGM, binding of the three swine isolates correlated well with SNA binding, which was consistent with the and swine isolates being generally specific for α2,6-Sia. The three strains analyzed bound a number of common glycan fractions, including glycans identified with chart identification nos. 64–67 and 85–89 (Fig. 3 and Dataset S2).

Human isolates.

We analyzed a vaccine strain, A/Brisbane/59/2007 (H1N1), seasonal strains A/Pennsylvania/08/08 (H1N1) and A/New York/55/2004 (H3N2), and three pandemic strains of H1N1 A/California/04/2009, A/Mexico/INDRE4487/2009, and A/Texas/15/2009 side-by-side on defined glycan microarrays of the CFG and the PL-SGM (Fig. 3 and Dataset S2). The vaccine strain A/Brisbane/59/2007 (H1N1), which had undergone multiple passages in eggs, had an extremely broad specificity, binding ∼100 of the 158 sialylated glycans on the CFG array, including 50 of the 88 glycans that have α2,3-linked sialic acid. Among the seasonal and pandemic strains analyzed, the seasonal strain A/Pennsylvania/08/08 (H1N1) had the broadest specificity, binding 55 glycans in the top 80 percentile ranking, which included only 11 glycans containing only α2,3–linked sialic acid. The other seasonal strain A/New York/55/2004 (H3N2) displayed the most restricted specificity on the CFG-defined array with binding to only 10 glycans considered as significant; all of which contained α2,6Sia, and in most cases, these terminated linear lactosamine and polylactosamine structures. Among the pandemic strains from 2009, A/Texas/15/2009 (H1N1) showed the broadest specificity, binding 38 sialylated glycans, all containing α2,6-linked sialic acid. Glycans bound by the other pandemic strains, A/California/04/2009 (H1N1) and A/Mexico/INDRE4487/2009 (H1N1), were also bound by A/Texas/15/2009. When the human isolates were analyzed on the PL-SGM, the vaccine strain A/Brisbane/59/2007 (H1N1) appeared to have the broadest specificity for the pig lung N-glycans, binding 28 of the 96 glycans printed on the PL-SGM (Fig. 3 and Dataset S2), which is consistent with the broad specificity of this strain on the defined CFG array. Similar to its behavior on the CFG array, the seasonal strain A/New York/55/2004 (H3N2) showed the most restricted specificity on the PL-SGM (Fig. 3 and Dataset S2), with only four glycans bound to significant levels. Thus, the general behavior of the human isolates on the PL-SGM was very similar to their behavior on the defined glycan array.

Some general observations from the CFG and the PL-SGM data can be made when comparing the isolates, especially because they are grouped by specificity. The avian isolates, which demonstrated a preference for terminal α2,3-Sia, preferentially bound to glycans displaying core fucosylation of N-glycans, especially A/Ruddy Turnstone/DE/650645/2002 (H1N9). For both avian viruses, residues of the same structure tend to bind with greater apparent affinity if a core fucose is present. However, the subset of viruses that bind terminal α2,6Sia do not show a preference for core fucosylated glycans. For example, the A/sw/Minnesota/02749/2009 (H1N1) virus binds two structures in the highest percentiles (100 and 96) that differ only in the addition of core fucose residues. For the mammalian viruses, the sialic acid linkage and composition of the terminal sugars seem to be more critical determinants of binding, because all glycans with these terminal characteristics are bound with similar affinity, regardless of chain length or whether they are branched or unbranched. This finding suggests that for the recognition of terminal α2,6-Sia, it is the extreme reducing end of the glycan structure that is important, and this may be attributable to the way the terminal sialic acid in the α2,6 conformation fits into the binding pocket on the HA. The avian viruses seem to rely more on core structure and modification for binding preference in comparison with the mammalian strains, which could be a consequence of the HA receptor interactions beyond the sialic acid binding pocket of HA.

Characterization of Glycans That May Function as Influenza Virus Receptors.

The N-glycans on the PL-SGM that were consistently bound at high levels by labeled viruses were considered candidates as natural receptors for influenza and chosen for further structural analysis. Glycans 34, 64, 67, and 88 showed relatively strong binding by human, swine, and avian influenza virus strains, whereas glycans 56, 81, and 83 were recognized by the avian virus strains and the human and swine strains, A/Brisbane/59/2007 (H1N1) and A/sw/Illinois/02860/2009 (H1N2), that showed a broader specificity of binding (Fig. 3 and Dataset S2). Therefore, the structures of these glycans were analyzed by MALDI-TOF, tandem mass (MS/MS) spectrometry, and the pattern of binding by lectins with defined specificities (SI Appendix, Figs. S3–S8, Datasets S1 and S3, SI Appendix, Table S5), to allow a prediction of glycan structure using the general approach of metadata-assisted glycan sequencing (MAGS) (52).

The glycans were analyzed in both positive and negative mode MALDI-TOF to check for sulfate groups in the glycans. A difference of m/z 78 (or 56 because of sodium adduct) indicated a sulfo group. The glycan structures were also analyzed by MS/MS to determine the backbone structures. In addition, the glycans were further treated with neuraminidase to demonstrate the sialic acid moiety and linkage, which is also easier for MS/MS characterization of glycan backbones. For example, when glycan 67 was treated with neuraminidase, its HPLC profile was shifted, indicating sialic acid removal (Fig. 4A). MS revealed glycan 67 has a molecular mass of 2,553.2 [M+Na]⁺, very close to calculated molecular mass of 2554.9 [M+Na]⁺, indicating a composition of five hexoses (Hex), four acetyl hexosamines (HexNAc), two sialic acids (Neu5Ac), and one fucose (Fuc) (Fuc₁NeuAc₂Hex₅HexNAc₄) (Fig. 4B). Together with the molecular weight change determined by MS (Fig. 4C), we propose two sialic acids in the glycan moiety. The MS/MS further confirmed the backbone structures (Fig. 4C). Consistent with the MS data, the glycan was bound by SNA and RCA-I, which is consistent with Siaα2,6Galβ1,4GlcNAc, whereas ConA binding is consistent with a biantennary N-glycan (not a triantennary complex N-glycan structure) and AAL binding with the presence of fucose. Taken together, these results indicate that glycan 67 is a biantennary complex-type N-glycan terminated with Siaα2,6Galβ1,4GlcNAc and containing a core fucose linked to the reducing end GlcNAc, as shown in Fig. 5. This structure is consistent with no binding by ECL (free terminal galactose) or MAL-I (α2,3-linked sialic acids). The MS analysis of glycan 64 predicts a biantennary complex-type N-glycan with no core fucose (Fig. 5) and a lectin binding pattern identical to glycan 67 with the exception of AAL, which is consistent with the absence of Fucose in this glycan. In addition, glycan 64 demonstrated a lectin-binding profile identical to the standard glycan 99; glycans 64, 67, and 99 all showed similar virus binding profiles.

Fig. 4.

Determination of the structure of fraction 67. (A) HPLC profile of fraction 67 before and after treatment of neuraminidase. (B) The MS spectra of fraction 67, in both positive (Upper) and negative (Lower) mode. (C) The MS (Upper) analysis of fraction 67 after treatment of neuraminidase and MS/MS (Lower) analysis of parent ion 1972.

Fig. 5.

The structures of selected N-glycans. The bound glycans were identified as bi- and triantennary complex-type N-glycans, as well as a sulfated biantennary structure with α2,6-linked sialic acids (Upper) or α2,3-linked (Lower) sialic acids. See Fig. 4 for glycan cartoon key.

Based on a similar approach, the other glycans were identified as bi- and triantennary complex-type N-glycans (Fig. 5), as well as a sulfated biantennary structure (glycan 83). The sialic acid linkages were identified from the lectin-binding profiles (SNA and MAL-I). All of the candidate virus-binding receptors terminated in α2,6-Sia, except glycans 56, 81, and 83, which terminated in α2,3-Sia based on lectin binding. Based on these predicted structures, comparisons with the structures present on the CFG array can be made. Glycans 64 and 67 have cognate structures on the CFG array; glycans 55, 318, 325, and 366 correspond to glycan 64 on the PL-SGM; and glycans 482 and 483 correspond to glycan 67 on the PL-SGM. The remaining structures from the PL-SGM that we have characterized are not represented on the CFG array. This finding confirms the novelty and significance of the shotgun array technique. To our knowledge, these data demonstrate for the first time the direct association of naturally occurring N-glycans from pig lung with binding of influenza viruses, supporting their function as potential receptors for these influenza viruses in vivo.

Studies on Virus Binding to O-Glycan and Glycolipids.

Although the above results define the ability of influenza viruses to recognize endogenous N-glycans, we also explored the potential of viruses to bind other classes of glycans, including O-glycans and glycolipids. Using the shotgun glycomics approach described previously (35), we prepared total O-glycans and glycolipids. These were derivatized with AEAB, separated by multidimensional HPLC, and printed on glycan microarrays. Although a wide variety of glycans were detected by specific lectin binding (SI Appendix, Figs. S9 and S11), as might be expected from previous studies of lung tissue glycomics of other species, no significant binding of influenza viruses was observed on shotgun microarrays of the O-glycans or glycolipid-derived glycans (SI Appendix, Figs. S10 and S12). Therefore, based on our available data, pig lung-derived N-glycans are recognized by all of the influenza viruses studied, but we observed no significant binding to pig lung-derived O-glycans or glycolipids.

Discussion

The data presented here provide fundamental contributions to the understanding of several separate but interrelated areas involving host–pathogen interactions. From a broad perspective, it represents, to our knowledge, the first interrogation of the pig lung glycome and opens up the field for investigating numerous economically important respiratory diseases of swine. Furthermore, the extension of the technology, reported here, to human tissue should pave the way for the understanding of various human respiratory diseases and promote the development of novel therapeutic intervention strategies. Our results also reveal novel insights into the recognition of natural glycan receptors in the pig lung by a diverse array of influenza viruses and show that binding does not necessarily follow basic assumptions that have been made regarding α2,3-Sia and α2,6-Sia linkage specificity for avian and human viruses, respectively.

Because swine are susceptible to both avian and human influenza viruses and are considered a possible intermediate host for the generation of pandemic human viruses, there has been significant interest in characterizing the glycome using glycomic profiling of cells and tissues by MS (34, 53). These studies reveal the presence of a diverse range of glycans terminating in Galα1-3Gal or sialic acid in both α2,3- and α2,6- linkages, where NeuAc was more prevalent than NeuGc. Glycan microarray analyses of the same viruses that were used to infect primary swine respiratory epithelial cells demonstrated the viruses preferred α2,6-linked sialic acid on the array, and infectivity studies showed a clear dependence of surface α2,6–linked sialic acid for influenza infection (34). These studies indicated that NeuAcα2,6-containing N-glycans are important for virus infection, but structural definition of the natural glycan receptors was predicted only by comparison of the similarity of the predicted glycan structures determined by MS analysis with the defined glycans on the CFG glycan microarray, and therefore these studies are correlative. In addition, glycans expressed by cultured cells isolated from organs may be misleading and not necessarily reflective of glycans in the native organ, because cell culture conditions have been known to select for changes in glycan structures (54, 55). Our approaches involve direct analysis of glycan structures in the relevant tissue. We analyzed two independent samples of pig lung tissue, observing glycans in the range of 1,500–5,000 Da, and the data are similar to the profiles reported previously for trachea, lung, and primary swine respiratory epithelial cells. Whereas the MS profile of primary swine respiratory epithelial cells observed glycans over an extended mass range of 1,500–8,000 Da (34), the large, multiantennary polylactosamine-containing glycans that constitute the high end of the mass range were present in vanishingly low amounts. We were unable to detect these glycans on the lung sample we analyzed, and they were not reported as present in trachea or lung by Sriwilaijaroen et al. (53); however, no testing of influenza binding to these glycans was performed in this study. The defined glycan microarray currently available from the CFG contains a variety of multiantennary glycans possessing long stretches of polylactosamine terminated in sialic acid, and numerous analyses of influenza virus binding to this array of defined glycans demonstrate strong virus binding to such structures. Based on these data, it has been suggested that such long sialylated polylactosamine structures may be important in productive virus infection (56, 57). Such conclusions, however, may be open to question, because the amounts of such structures actually identifiable in the tissues appears to be extremely low or undetectable. As we have demonstrated, if sufficient material is available to undergo detailed structural analysis, the shotgun glycan microarray approach makes it possible to identify potential receptors based on their presence in natural tissue and their ability to participate in functional binding.

A significant body of work has accumulated from many investigators on analyses of influenza A virus binding to defined sialylated glycans on glycan microarrays and despite this approach, we still have not defined the exact structure of a glycan receptor that functions in productive infection. These data were recently documented in a multidisciplinary study of influenza A virus–glycan interactions using a glycomics approach (21). In this extensive study, the N-glycan and O-glycan composition of human lung and bronchus were analyzed by MS. A wide spectrum of α2,3-Sia– and α2,6-Sia–terminated glycans were found in both the lung and the bronchus. These results indicate that alveolar tropism by influenza viruses appears to involve both α2,3-Sia– and α2,6-Sia–terminated glycans and emphasizes the complexity of this system where the cell surface glycans, as well as glycoconjugates of the respiratory tract fluid probably, play important but undefined roles. The MS data allowed for structural predictions that were used in combination with influenza A virus-binding data from four published glycan arrays comprised of defined glycans to determine whether they could predict replication of human, avian, and swine viruses in human ex vivo respiratory tract tissues. The CFG defined glycan microarray was the most comprehensive of the four evaluated. Although the CFG array possessed the greatest diversity of sialylated glycans, the influenza A binding data from this array was not predictive of productive virus replication in human respiratory tract tissue. The study concluded that more comprehensive and focused arrays need to be developed to evaluate emerging influenza A viruses. It is tempting to conclude that the solution to this dilemma is simply to synthesize the glycans corresponding to the sialylated glycans predicted by glycomics analyses. However, the glycan structures based on profiling are predicted structures, and details of all of the structures are not available. Furthermore, synthesis of large N-glycans is an extremely difficult task, and synthesis of all possible glycans predicted by MS analysis of the human respiratory tract tissues is currently not possible. Thus, the shotgun glycomics approach, which will obtain and identify the glycans from the natural source of tissue or glycoconjugate, has a high probability of presenting cellular glycans that function as viral receptors.

With regard to the specific binding properties of influenza viruses, our data are generally consistent with the well-documented observations that some strains bind preferentially to either α2,3-Sia or α2,6-Sia and that these specificities can provide useful indicators for potential involvement in host species specificity. However, our results also indicate that influenza virus receptor recognition is much more complicated than might be ascertained from such basic assumptions. One important observation is that virus binding profiles do not directly correlate with the recognition profiles determined for the lectins SNA and MAL-I, which are often used to assay for the abundance of specific “receptors” of α2,6-Sia or α2,3-Sia linkages, respectively (17, 19, 20, 22, 58). Whereas some viruses, such as the vaccine strain A/Brisbane/59/2007 (H1N1) and the low-pathogenicity avian isolates A/Ruddy Turnstone/650625/2002 (H6N1) and A/Ruddy Turnstone/650645/2002 (H1N9), bound to glycans that broadly reflected the profile observed with the specific lectins, other viruses, such as the pdmH1N1 strains A/California/04/2009, A/Mexico/INDRE4487/2009, and A/Texas/15/2009 and the seasonal strains A/Pennsylvania/08/2008 (H1N1) and A/New York/55/2004 (H3N2), bound to highly specific subsets of glycans. Interestingly, for a number of virus strains, significant binding was observed to glycans recognized by neither SNA or MAL-I. In other examples, viruses bound efficiently to both α2,3-Sia– and α2,6-Sia–linked glycans. A/swine/Minnesota/02719/2009 (H3N2) and A/swine/Illinois/02860/2009 (H1N2) each bind the same three glycans that are recognized as α2,3-Sia and α2,6-Sia. Similarly, three glycans of both specificities are bound by A/Brisbane/59/2007 (H1N1), A/California/04/2009 (H1N1), and A/New York/55/2004 (H3N2). This dual specificity may have been expected for the vaccine strain A/Brisbane/59/2007 (H1N1), because it represents a human strain that had been subsequently passaged in embryonated chicken eggs, where recognition of α2,3-Sia linkages may have been selected but was unexpected for other viruses that bound the same glycans or for the two avian strains A/Ruddy Turnstone/DE/650645/2002 (H1N9) and A/Ruddy Turnstone/650625/2002 (H6N1). Although these avian strains showed the expected binding to α2,3-Sia glycans that were not recognized efficiently by human or swine strains, they also bound to several of the α2,6-Sia–containing glycans. The swine viruses A/sw/Minnesota/02749/2009 (H1N1), A/sw/Minnesota/02719/2009 (H3N2), and A/sw/Illinois/02860/2009 (H1N2) were all isolated in 2009 but displayed broad differences with respect to the number of glycans recognized, with the A/sw/Minnesota/02749/2009 (H1N1) virus showing a much more restricted binding profile. Although these swine viruses were isolated in the US Midwest in the same year, they are different HA and NA subtypes. Another illuminating observation relates to the valency of receptors recognized, because some of the more highly specific strains were more selective for di- and triantennary glycans. Interestingly, the 2009 human pandemic strains known generally to have low binding activity (50, 59), were more selective for the multivalent receptors.

The glycans that bound virus and generated a ranking of greater than 40% for at least one of the viruses interrogated are shown in Dataset S3. The glycans selected for further analysis were glycans 34, 56, 64, 67, 81, 83, and 88, and their structures are shown in Fig. 5. The characterized glycans do have some similarities, which suggests that these features might be relevant determinants of binding. Each structure is branched, either biantennary or triantennary, and of the eight structures characterized, seven contain a core fucose modification. The motif of terminal Sia-Gal-GlcNac is represented in all of the structures that have been characterized which is in accordance with general observations of influenza binding. These structures were chosen because of their abundance, purity, and relative ease of characterization. Each is a principal binder for at least one and often many of the viruses tested, but they are not necessarily the highest binding structure. Further structural characterization of the components of the PL-SGM will reveal the true relevance of these complex N-glycan structures with respect to influenza binding.

The observation that swine and human isolates recognize a number of the same glycans is consistent with our knowledge of the swine as a suitable host for a range of influenza viruses, and this is supported by the observation that avian isolates examined here also recognized a range of both α2,3-Sia– or α2,6-Sia–linked receptors. The extension of the pig lung glycan array technology reported here to human respiratory tract tissues, and to the tissues of the intestinal tract of avian species, will be required to identify whether specific receptors exist that are critical determinants of host range.

Our current understanding of influenza binding specificity has been based largely on studies of binding to artificial receptor analogs of defined α2,3-Sia or α2,6-Sia linkage or to heterogeneous collections of glycans present on cell surfaces characterized only by reactivity with specific lectins. Such studies have proven useful for broad generalizations of species specificity, by relating them to the amino acid composition, and structural aspects of the HA-binding site using viruses isolated from different hosts. However, a striking feature of these results presented here relates to distinct differences in the binding profiles detected, even among strains that would be broadly classified in the same category based on overall α2,3-Sia or α2,6-Sia preference. These distinctions relate to glycan backbone, modifications, and valency of the receptor species and highlight a level of complexity that is not easily explained based on our current knowledge of HA structures. When considering productive virus attachment in the context of a natural host, the situation becomes even more complicated, because factors relating to NA activity and specificity, relative ratios and distribution of HA and NA glycoproteins on viral surfaces, and virion morphology may all play a role. The mechanisms by which individual virus strains modulate their detailed selectivity for receptors, and the implications for host range and pathogenicity, are not currently appreciated. However, our current studies on swine tissue suggest that it will require a much broader understanding of the identity, density, and distribution of natural endogenous receptors at sites of infection in many of the natural hosts for influenza viruses.

Experimental Methods

Materials.

Pig lung was purchased from Lampire Biological Products or obtained from germ-free animals from the College of Veterinary Medicine, University of Minnesota, and stored at −80 °C until use. Euthanasia and tissue harvest was approved by the University of Minnesota Institutional Animal Care and Use Committee, was conducted in compliance with the Animal Welfare Act, and adhered to principles stated in the Guide for the Care and Use of Laboratory Animals (60) (protocol 1011A92934). All chemicals were purchased from Sigma-Aldrich and were used without further purification. HPLC solvents were purchased from Fisher Scientific. An Ultraflex-II TOF/TOF system (Bruker Daltonics) was used for MALDI-TOF MS analysis. Both reflective positive and reflective negative modes were used as indicated in the figures. The 2,5-dihydroxybenzoic acid [5 mg⋅mL⁻¹ in 50% (vol/vol) acetonitrile, 0.1% TFA] was freshly prepared as the matrix and 0.5 μL of matrix solution was spotted on an Anchorchip target plate (200 μm or 400 μm) and air-dried. Then 0.5 μL of sample solution was applied and air-dried. An ozone generator (model OZV-8; Ozone Solutions) was used to generate ozone from high purity oxygen.

Isolation of N-Linked Glycans.

Frozen pig lung (free of trachea) was homogenized at 4 °C and adjusted to a chloroform/methanol/water ratio of 4:8:3 (vol/vol/vol). The soluble fraction was reserved as a source of glycolipids and the insoluble material, collected by centrifugation, was resuspended in in 8 M GuHCl in 0.2 M Tris (pH 8.2) and subsequently reduced by DTT and alkylated by iodoacetamide. Denatured material was dialyzed against water overnight and lyophilized. The lyophilized proteins were digested with trypsin in 50 mM phosphate buffer (pH 8.2) to generate glycopeptides that are susceptible to digestion with Peptide N-Glycosidase F (PNGase F; New England Biolabs). After PNGase F digestion, the released N-glycans were obtained in the run-through of C-18 cartridges (Waters) and collected by subsequent application to and elution from a Carbograph cartridge (Alltech), and subsequently the free glycans were derivatized with AEAB as previously described (38). The bifunctional fluorescent AEAB tag permits quantification of total glycans and provides a primary amino group for coupling to NHS-activated glass surfaces when printed as microarrays.

Isolation of O-Glycans and Glycolipids.

O-glycans were isolated following the nonreductive elimination procedure using ammonium hydroxide/ammonium carbonate developed by Huang et al. (61). The free reducing glycans were conjugated with AEAB as described for N-glycans. Free glycans from glycosphingolipids were isolated following the ozone-mediated nonreductive delipidation (62). The free reducing glycans were conjugated with AEAB as described for N-glycans. One-dimensional HPLC using a normal phase column was carried out for both O-glycans and glycans from glycolipids. Fractions were collected, lyophilized, quantified based on fluorescence, and printed without 2D HPLC separation. No apparent binding was observed for these microarrays.

Permethylation of Glycans.

For structural profiling by MALDI-TOF analysis, the N-glycans were permethylated using sodium hydroxide and iodomethane in DMSO (37) before derivatization with AEAB. The permethylation reaction was quenched with water after 1 h, and the samples were extracted with chloroform. The organic phase was washed three times by ddH₂O and gently dried with nitrogen gas under a hood and redissolved in 1:1 methanol:water. The samples were loaded onto C-18 cartridge and eluted stepwise with water and 15%, 35%, 50%, and 75% aqueous acetonitrile. Permethylated glycans were usually present in the 35%, 50%, and 75% acetonitrile fractions. The glycans were annotated with the help of the GlycoWorkbench tool.

HPLC Separation of AEAB.

An HPLC CBM-20A system (Shimadzu), coupled with a UV-light detector (SPD-20A) and a fluorescence detector (RF-10AXL), was used for HPLC analysis and separation of AEAB-conjugated glycans; 2D HPLC separation, using normal-phase and reverse-phase columns as previously described (38) provided multiple fractions that were largely homogeneous. The quantified fractions obtained after 2D HPLC were then lyophilized and adjusted to 200 µM with deionized water and stored frozen as the PL-TGL.

Viruses and Cells.

Virus stocks representing swine, human, and avian isolates were selected based on their availability and the relevance of the strain with respect to circulation in a population. The virus stocks were grown in Madin–Darby canine kidney (MDCK) cells that were maintained using DMEM supplemented with 5% FBS and penicillin–streptomycin. Briefly, 85–90% confluent MDCK cells grown in T175 flasks were washed twice with PBS, overlaid with 10 mL of serum-free DMEM supplemented with 1 μg/mL tosylsulfonyl phenylalanyl chloromethyl ketone (TPCK) trypsin, and infected with a 1:1,000 or 1:10,000 dilution of the original virus stock. Cells were incubated with rocking for 1 h at ambient temperature, the inoculum was removed, and 30 mL of serum-free DMEM supplemented with 1 μg/mL TPCK trypsin was added. The flasks were incubated at 37 °C for 2–3 d until monolayers were 80–90% destroyed. Virus was then harvested in the infected cell supernatant and frozen at −80 °C for subsequent purification. The stocks of all viruses used in these studies were resequenced after the limited amplification in MDCK cells before labeling for glycan array analysis. We detected no sequence changes and no evidence for observable heterogeneity.

Purification of Virus Strains.

Preparations of each virus were done so that each would yield enough volume for multiple experiments so that essentially the same preparation could be used on each array at the same time. For virus preparations, infected cell supernatants were clarified by low speed centrifugation, and then pelleted through a 25% sucrose cushion in NTE buffer (100 mM NaCl, 10 mM Tris, 1 mM EDTA). Purified viruses were pelleted by centrifugation in a Beckman Coulter SW32 rotor at 28,000 rpm for 2 h, resuspended in 2 mL of PBS buffer, aliquoted, and frozen at −80 °C. Frozen, purified virus strains were later thawed, and the hemagglutination units and plaque forming units (plaque forming units per milliliter) titers were determined using standard techniques.

Agglutination of Erythrocytes.

Chicken erythrocytes were acquired from Lampire Biologicals as whole-blood preparations. Whole-blood preparations were washed two times with 1× PBS and diluted to 0.5% for hemagglutination experiments, which were performed using standard techniques. Briefly, 0.5% erythrocyte preparations were added to twofold serial dilutions of 50 μL of virus stock for 1 h to determine the HA titer. This procedure was done before and after labeling the virus with Alexa Fluor 488 to monitor the HA titer during the labeling process. Infectious titers of virus stocks were also determined by plaque assay on MDCK cell monolayers.

Fluorescently Labeled Viruses.

Binding of fluorescently labeled influenza virus was performed as previously described (50, 63, 64). Briefly, 200 μL of virus was incubated with 25 μg of Alexa Fluor 488 (Invitrogen) in 1 M NaHCO₃ (pH 9.0) for 1 h at room temperature. Labeled viruses were dialyzed against PBS using an M_r 7,000 molecular weight cutoff Slide-A-Lyzer minidialysis unit (Thermo Scientific) overnight at 4 °C. In all cases, labeled viruses were used in experiments the following day.

Microarray Preparation and Analysis.

NHS-activated slides (Nexterion H) were purchased from SCHOTT North America. Corning Epoxy slides were purchased from Corning Incorporated Life Sciences. Noncontact printing was performed using a Piezoarray printer (Perkin-Elmer) as previously described (65). All samples were printed within 10% variation of 0.33 nL at a concentration of 100 µM in phosphate buffer (300 mM sodium phosphates, pH 8.5). Before assay, the slides were rehydrated for 5 min in TSM buffer [20 mM Tris⋅HCl, 150 mM sodium chloride (NaCl), 0.2 mM calcium chloride (CaCl₂), and 0.2 mM magnesium chloride (MgCl₂)], and the analyses were carried out for lectins and viruses as previously described for 1 h at 4 °C to prevent neuraminidase activity (63, 65). Biotinylated lectins (Vector Labs) were used as controls in the binding assay and the bound lectins were detected by a secondary incubation with cyanine 5–streptavidin (5 μg⋅mL⁻¹; Invitrogen). For multipanel experiments on a single slide, the array was constructed according to the dimension of a standard 16-chamber adaptor. The adaptor was applied to the slide to separate a single slide into 16 chambers sealed from each other during the assay. The slides were scanned with a ProScanArray microarray scanner (Perkin-Elmer) equipped with four lasers covering an excitation range from 488 to 637 nm, and the scanned images were processed to Excel spreadsheets as previously described (63, 65). For cyanine 5 fluorescence, 649 nm (excitation) and 670 nm (emission) were used. For Alexa Fluor 488 fluorescence, 495 nm (excitation) and 519 nm (emission) were used.