Note our study used pseudoviruses and conditionally replicative virions to ensure biosafety, and reports deep mutational scanning only for HA usage of tufted duck MHC-II to limit any information hazard concerns.

We also showed H7 HA binds tufted duck MHC-II similarly to H5 HA, & some H1, H2, H3, & H9 HAs also can use avian or human MHC-II. But patterns vary among strains. For instance, an avian influenza HA and the 1918 HA can use tufted duck MHC-II, but later human strains cannot.

Structure & deep mutational scanning suggest identity of peptide bound to MHC-II could influence interaction of HA & MHC-II. Also, HA binding would likely block ability of MHC-II to interact with T-cell (perhaps analogous to how EBV gp42 can bind to MHC-II to block T cell activation).

So although structure only 4.8 A, it is corroborated by deep mutational scanning of both HA and MHC-II showing that sites in both proteins that affect binding are at structural interface. (Sites where mutations decrease binding are red in structure below)

To validate structure, we used inverted pseudotyping deep mutational scanning to measure how mutations to tufted duck MHC-II affect binding to H5 HA. Most mutations with big impact in alpha chain, but beta-chain mutations near peptide-binding groove also have effect.

To determine structure, we produced H5 HA protein w mutations that increased binding to tufted duck MHC-II. This HA increased fraction of particles bound to MHC-II in ns-EM, and we were able to use it solve cryo-EM structure of H5 HA bound to tufted duck MHC-II.

The mutations that reduced MHC-II entry clustered in a region on HA head defining the MHC-II binding surface. We also directly measured how HA mutations affect binding to tufted duck MHC-II, and identified same binding surface.

To understand how HA interacts w MHC-II, we measured how all H5 HA mutations affect pseudovirus entry via sialic acid or tufted duck MHC-II. Identified loss-of-function mutants that could only use MHC-II or sialic acid.

We tested 80 H5 HAs: most but not all could enter cells via tufted duck & to lesser extent human MHC-II. Note MHC-II highly variable within and between species. See dms-vep.org/Flu-H5N1-Ame... for interactive version of below plot.

We first measured ability of two H5 HAs to enter cells via sialic acid or MHC-II. As shown below, both HAs could use tufted duck & to lesser extent human MHC-II. [Note: experiments used pseudoviruses, which can only undergo single round of cell entry, providing safe way to study HA]

We also defined how F mutations affect neutralization by a panel of monoclonal antibodies. This allowed us to quantify the resilience of different antibodies to escape, and predict which antibodies also neutralize the related Hendra virus.

A strategy for vaccines is to stabilize F in pre-fusion conformation. We identified sites where mutations to proline (which blocks helix formation) are disfavored. This identifies new candidate mutations for stabilizing F vaccine immunogens.

For this study, we used pseudoviruses that can only undergo a single round of cell entry (& so are not human pathogens) to measure how mutations to F affect its fusion function. We found F is more functionally constrained than the other Nipah surface protein, RBP.

For instance, in interactive plot below I've moused over to highlight serum from an individual who mostly has high titers, but has dramatically reduced titers just to strains with mutations at site 135. There is no way to represent that sort of thing w just medians and ranges across sera.

For H1N1 influenza, a new subclade (D.3.1.1) has also recently spread to become dominant, and our data show that this new subclade has reduced neutralization by human sera See jbloomlab.github.io/flu-seqneut-... for interactive version of below plot

The mutations that further reduce neutralization of subclade K are in antigenic regions D & E, which were less mutated in parent subclade K See below from recent @scottehensley.bsky.social preprint (doi.org/10.64898/202...) & stay tuned for study from their group that explains this observation

For vaccine update decisions, we care about what is NEXT. Here our data help by showing that within subclade K strains there are already new subvariants w further reduced neutralization. These subvariants have additional mutations as shown below & interactively at nextstrain.org/community/jb...

Probably because of these lower titers, subclade K has rapidly become dominant among H3N2, rising from <1% to 95% frequency in ~9 months. See below image from this Nextstrain link (nextstrain.org/seasonal-flu...).

Resulting datasets are very rich. Below are H3N2 data (also under first post in this thread). There is extreme variability among human sera; the median serum has ~1.5-2-fold lower neutralization of subclade K. See jbloomlab.github.io/flu-seqneut-... to explore interactive plot

Specifically, we first assembled a set of 57 H3N2 and 34 H1N1 strains that largely cover the current diversity of human seasonal influenza (see image below). We then measured neutralization of all 91 strains against 302 sera from humans of a range of ages (0 to 103 years) and geographic locations.

Above visualizations just scratch surface of data: there is tremendous heterogeneity across sera from different individuals not easily summarized by median/mean. Indeed, we previously found this heterogeneity may be important for influenza evolution: elifesciences.org/reviewed-pre...

We then measured how 188 human sera recently collected at four different sites neutralized all 140 influenza strains in library. Titers are summarized below; can be examined interactively at jbloomlab.github.io/flu-seqneut-... & jbloomlab.github.io/flu-seqneut-...

In spring of 2025, we designed library of naturally occurring human seasonal influenza strains that represented diversity of available sequences at that time; this library continues to cover most sequenced diversity of H3N2 and H1N1 hemagglutinin today.

To do this, we used sequencing-based neutralization assays that measure many neutralization curves simultaneously (journals.asm.org/doi/10.1128/... & elifesciences.org/reviewed-pre...) Approach enabled one grad student (@ckikawa.bsky.social) to measure ~26,000 neutralization curves in ~5 months.

But because it takes time to perform experiments, measurement of how current strains are neutralized by human serum antibodies can lag timeline for vaccine strain selection. Our goal was to use new approach to characterize human antibody landscape at scale in near real-time.

Data in interactive form at dms-vep.org/CHIKV-181-25... Thanks to Xiaohui Ju for leading study Special thanks to @msdiamondlab.bsky.social for help Also Will Hannon, Caelan Radford, Brendan Larsen, Daved Fremont, Ofer Zimmerman, Tomasz Kaszuba, Chris Nelson, Israel Baltazar-Perez, Samantha Nelson

After using pseudoviruses & reporter particles to show mutations *loss* of function, we engineered into Chikungunya virus: mutants lost ability to infect human or mosquito cells. So we reduced natural tropism for both human & mosquito cells to just one type of cell.

We next used non-replicative single-cycle alphavirus reporter particles (which provide another safe way to study mutations) to validate that mutations identified in deep mutational scanning indeed specifically impaired entry in human or mosquito cells only.

Sites where mutations specifically impair entry in 293T-MXRA8 cells mostly at MXRA8 binding interface. We also find sites where mutations specifically impair entry in C6/36 cells. Although mosquito receptor unknown, we hypothesize these sites at its binding interface.

Most mutations similarly affect entry in all three cells, but some have cell-specific effects. For instance, mutations at E2 site 119 are generally tolerated in C6/36 and 293T-TIM1 cells, but deleterious in 293T-MXRA8 cells. (See dms-vep.org/CHIKV-181-25... for interactive plot.)