Do novel H5N1 hemagglutinin variants from a human case bind sialosides poorly?

Two assays, one consistent binding loss

The paper asks whether the HA variants found in the BC24 patient still attach to the sialic acid receptors influenza uses to enter cells. Two complementary binding assays answered this: a glycan microarray printed many defined glycans on spots and read background-subtracted fluorescence at 532 nm, while an ELISA measured absorbance at 450 nm for HA binding to immobilised biantennary α2,3- and α2,6-linked sialosides (3SLN-N, 3SLN3-N, 6SLN-N, 6SLN3-N).

The comparison set matters. VN04 HA bound exclusively to α2,3-linked sialosides on both N-linked and O-linked arrays, to the limit of detection. MI24 HA did the same but showed a mild preference for sialyl Lewis X, consistent with the broader binding reported for clade 2.3.4.4 HA proteins. BC24 HA, by contrast, showed no binding above the microarray threshold of fifty-fold background to any α2,3-linked sialoside or any glycan on the array, and no detectable binding to the four sialosides in the ELISA.

That is a statement about assay sensitivity, not proof of zero affinity. The authors note BC24 HA may bind biologically relevant sialosides at low affinity below the detection limit of ELISA and microarray, possibly compensated by avidity in the fusion assay, or may use alternative receptors not tested here.

The receptor pocket loses its own glycan

Cryo-EM structures of pre-fusion HA ectodomains, expressed and purified identically for VN04, MI24 and BC24, gave the mechanistic entry point. All three trimers were globally similar, with RMSD of 0.441 Å for BC24 versus VN04 and 1.095 Å for BC24 versus MI24, and fusion peptides stayed sequestered in the helix C bundle as expected at pH 8. The E190D and Q226H mutations did not produce substantial structural change near or far from those residues.

The striking difference was at the receptor-binding site. Clear density for the N169 auto-glycan — modelled as GlcNAc-GlcNAc-Man-Man-GlcNAc-Gal-Sia, with the terminal linkage in the α2,3 configuration — was present in VN04 and MI24 HA but absent in BC24 HA. This auto-glycan normally occupies the sialic acid binding site and is displaced by host-cell sialosides during infection, so its absence suggested greatly reduced affinity for α2,3-linked sialosides.

The authors checked whether BC24 HA simply fails to glycosylate N169. Density from N-linked sites at N21, N33, N158 (VN04 only), N169, N289 and N483 was visible across the three proteins, and the first portion at N33, N169, N289 and N483 could be modelled. Because the same expression system was used and MI24 and BC24 ectodomains are closely related, the authors argue the missing pocket density reflects reduced glycan affinity rather than absent glycosylation — while conceding that a lack of auto-glycan binding may not matter for productive receptor engagement.

Residue geometry explains the lost contacts

Model-based measurements point to the two mutated positions. In VN04 and MI24 HA, the E190 carboxylate sits 2.9 and 2.8 Å from the Sia-1 C9 hydroxyl, and the Q226 amide sits 3.3 and 3.2 Å from the Gal-2 C4 hydroxyl; the MI24 Q226 amide is also 3.2 Å from the glycosidic oxygen linking Gal-2 and Sia-1. These are plausible hydrogen-bonding distances for holding the α2,3-linked auto-glycan.

In BC24 HA, with no ligand bound, the D190 carboxylate — one methylene shorter than glutamate — lies 4.9 Å from where the MI24-bound Sia-1 aligns. H226 pivots 1.1 Å away from the MI24-bound Gal-2 C4 hydroxyl position and 0.9 Å toward the C1 carboxylate. The receptor-binding site backbone itself is nearly unchanged, with RMSD of 0.448 Å between VN04 and BC24 and 0.378 Å between MI24 and BC24 over the binding-site residues. The interpretation is therefore about side-chain contacts, not a reshaped pocket.

Single and double mutants share the phenotype

The ELISA mutagenesis separates the two changes. Starting from MI24 HA, the E190D single mutant, the Q226H single mutant and the E190D+Q226H double mutant each showed reduced binding to both α2,3- and α2,6-linked sialosides. The reverse experiment on BC24 HA — D190E, H226Q, and D190E+H226Q — behaved the same way, with BC24 wild type showing no detectable binding.

For the four representative sialosides tested, each single mutant and the double mutant produced a similar reduced-binding phenotype. The authors conclude that receptor binding of the mutant viruses in this case does not depend on the specific combination of E190D and Q226H. Because single mutants are expected to have existed in the patient, the result implies multiple virus populations shared the diminished-binding phenotype rather than one double-mutant lineage.

Weak binders still fuse membranes

Binding assays do not show whether HA can do its job. The authors expressed full-length, untagged BC24 HA in A549 alveolar carcinoma cells and triggered fusion by acidification, reading syncytia formation and a NanoBiT luciferase signal normalised to VN04 HA within each biological replicate. A549 cells have previously been shown to display both α2,3- and α2,6-linked sialosides on their surface.

BC24 HA drove fusion, but at lower levels than VN04 and MI24 HA, with significance values against mock-transfected controls of p = 1.1541×10⁻¹⁰, 5.8429×10⁻⁶ and 1.4048×10⁻⁶ reported for the three comparisons. BC24 HA fusion was also statistically significantly greater than the commonly used Y98F binding-deficient mutant, although the signal ranges overlapped. Since proliferation has been reported with Y98F depending on cell type, the authors read the reduced-binding minority variants as functional.

The authors flag that A549 cells do not perfectly imitate real lung tissue, and that BC24 HA may bind biologically relevant sialosides below assay detection but be partially rescued by avidity in the fusion assay, or may engage untested alternative receptors.

No evidence for a receptor-specificity switch

A natural hypothesis for a human H5N1 case is that HA acquired α2,6 preference, the linkage more common in the human upper airway. The data do not support that here. BC24 HA bound neither α2,3- nor α2,6-linked sialosides detectably, and each mutation reduced binding to both types rather than trading one for the other. The authors state plainly that their findings do not support a switch in receptor specificity from α2,3- to α2,6-linked sialosides.

What remains is a variant that is worse at binding the canonical receptors in both configurations, yet retains fusion competence. The paper's proposed explanation is spatial rather than receptor-driven: because sequences came from a lower respiratory tract sample, the authors hypothesise that severely reduced binding affinity, at a level still compatible with fusion, may have enriched these variants in distal airways by helping the virus avoid receptor decoys or other inhibitory factors there. That is explicitly a hypothesis, not a measured outcome.

Patient sera and a broadly neutralising antibody still bind

Reduced receptor binding did not mean reduced immune recognition. Post-infection sera from the BC24 patient, sampled at four independent time points within 1 to 4 days of a negative serum Flu A result, showed strong binding to BC24 HA and to HA proteins from other influenza subtypes. The variant HA is therefore still antigenically recognisable by the patient's own antibody response.

The broadly neutralising antibody MEDI8852, which targets a stem epitope conserved across most HA subtypes, also neutralised BC24 HA potently. In NanoBiT luciferase fusion assays, estimated IC50 values were 25.3 nM for VN04 HA and 31.9 nM for MI24 HA, and BC24 was neutralised at least as effectively as both. Biolayer interferometry with MEDI8852 Fab confirmed specific binding in the nanomolar range for all three HA proteins. Because the stem epitope is distal to the receptor-binding site, these experiments used HA carrying an additional Y107F mutation to reduce receptor binding and improve yields.

Minority variants complicate the severity picture

The paper's most consequential observation is population-level. Up to a third of sequence reads in the BC24 case carried E190D or Q226H, and each mutation independently and together severely reduced sialoside binding. A substantial minority of the virus population therefore carried a weakly binding HA phenotype, and the authors describe this as revealing further complexity in the factors that may be present in severe avian influenza infection.

That complexity cuts against using receptor-binding strength alone as a severity proxy. The authors contrast cases: a previously described infection was limited to conjunctivitis, whereas the VN04 and BC24 cases were respectively fatal and severe while showing weaker receptor binding. Many factors shape disease severity, and the paper offers its spatial-enrichment idea as one potential contributor rather than a demonstrated cause.

The practical priority this paper sets is to characterise HA variants as a mixed population, not a single consensus sequence, and to test binding and fusion separately. A variant can fall below the detection limit of glycan microarray and ELISA for both α2,3- and α2,6-linked sialosides and still mediate fusion in alveolar cells — so a negative binding result should be read alongside fusion competence, the specific glycans assayed, and the variant's frequency in the patient, not as evidence that the virus cannot enter or spread.

Sources used in this answer