Deep mutational scanning predicts pandemic influenza sequences

From the Bedford and Bloom labs, Vaccine and Infectious Disease and Basic Sciences Divisions

Viruses that exclusively replicate in a certain host species can acquire the ability to infect a distinct species. Rapid introduction of a new viral strain into the human population can lead to pandemic viral transmission in the immunologically naïve recipient species, and this adaptation process can occur in avian influenza strains that gain mutations sufficient to support replication in human cells. This phenomenon is typically studied retrospectively in the lab by passaging virus through various host species cell lines and screening for mutations that allow growth in non-reservoir species, or by observing mutations that occur in nature. Previous work has revealed that influenza adaptation to mammalian hosts specifically requires mutations in viral polymerase to allow transcription and viral genome replication. The PB2 subunit of avian influenza RNA polymerase functions poorly in human host cells, likely due to an unproductive interaction with host cellular machinery. However, PB2 sometimes acquires mutations that result in improved function between PB2 in human cells, so the sites that acquire repeated mutations have become focal points for studying mammalian adaptation of influenza. However, research focusing solely on the mutations that have thus far arisen naturally may overlook potential, distinct mechanisms of zoonosis. To prospectively expand analysis to potential pandemic-causing influenza mutations, Shirleen Soh of the Bloom lab (Basic Sciences Division), along with Louise Moncla of the Bedford lab (Vaccine and Infectious Disease Division) mapped each possible single amino acid mutation in avian influenza PB2, identifying many novel human-adaptive mutations. They recently published their work in eLife.

The authors initially selected a PB2 protein from a representative strain of avian influenza and performed deep mutational scanning, mutagenizing all codons to create individual mutant viruses that spanned nearly all 14,421 possible amino acid combinations. They then passaged each strain through both human and avian cell lines and deep sequenced each viral library before and after infection to identify non-synonymous mutations that arose during adaptation to human cells. To quantify the benefit of a given mutation in human cells, the authors employed differential selection, a metric that computes relative enrichment of a mutation relative to the wildtype in human versus avian cells and indicates whether a mutation is favorable in adaptation to human cells. They then calculated the mutation effect, or the quantification of the relative benefit of a mutation compared to wildtype, and together these metrics identified the mutations that are most beneficial for influenza to acquire the ability to replicate in human cells. Of these hits, the authors pursued experimental validation of their top 18 putative human-adaptive mutations, only one of which had been previously identified. They also included several avian-adaptive and both human-and-avian-adaptive mutations for comparison.

Schematic of pipeline for deep mutational scanning and differential selection in human and avian influenza PB2.
Schematic of pipeline for deep mutational scanning and differential selection in human and avian influenza PB2. Figure from publication.

To functionally confirm the functional capacity of these putative human-adaptive mutations to increase polymerase function in influenza, they performed a minigenome assay to quantify the transcriptional output of each mutant viral RNA across two human cell lines, and found that the majority of human-adaptive mutations identified in the deep mutational scanning did in fact increase polymerase activity, demonstrating that their screen had identified novel human-adaptive mutations in avian PB2. To expand their mutation characterization, the authors also sought to identify mutants that increase viral growth through means other than polymerase by performing a competition assay, coinfecting human and avian cells with both mutant influenza strains. They compared the resulting ratios of mutant: wildtype and human: avian viral RNA after 48 hours, finding that almost all identified mutants increased viral growth, including those mutations that had not increased viral transcription, suggesting that human-adaptive influenza adaptations may increase viral replication in polymerase-independent ways that have not yet been characterized. Together, these experiments showed that human-adaptive mutations with both known and unknown functions in avian influenza can be prospectively predicted.

Finding that not all putative human-adaptive mutations directly affect polymerase activity, the authors hypothesized that these additional polymerase-independent mutations may increase adaptation to human cells by facilitating PB2 interactions with host cell factors. To test this, the authors interrogated the location where virus-human cell interactions may occur by mapping the top human-adaptive mutations onto the structure of PB2. They found that most clustered on the surface of PB2 near known regions of virus-host interaction, further suggesting that human-adaptive mutations may function by improving contact between virus and host cell. Finally, to confirm that their experimental measurements could accurately predict real avian-human influenza transmission events, the authors constructed phylogenic trees with historical H7N9 PB2 influenza sequences, assigning each mutation as avian or human based on its origin. They applied differential selection to these historical mutations and identified both known and novel human-adaptive mutations. These experiments found that human-adaptive mutations in nature may function through previously unstudied mechanisms and validated this prospective high-throughput experimental and computational approach as an accurate predictor of empirical human-adaptive influenza mutations.

The authors speculate on the implications of these studies, highlighting that deep mutational scanning of influenza human-adaptive mutations revealed not only naturally observed mutations but also mutations that have yet to occur in nature. They hypothesize that some putative mutations may be inaccessible by single nucleotide changes in existing sequences. Therefore, to predict pandemic-causing mutations that realistically may arise in nature, these deep mutational scanning data can be combined with studies of evolutionary accessibility. Furthermore, they posit that these mutational surveys of PB2 can be applied to more general characterization of its structure and function more generally, since their complete mapping of amino acid substitutions highlight individual sites’ tolerance for alternate residues. The authors stress that this approach is high-throughput and allows for rapid interpretation of real-time mutations in circulating influenza strains, which they are actively pursuing. Moncla explained that she and Soh “are working together to incorporate this data onto the avian flu build of so that we can visualize whether new, emerging strains of avian influenza harbor human-adapting polymerase mutations. That's something I'm really excited about." Soh added that “in the future, we hope to identify what these host factors are that influenza co-opts for its replication."

This work was supported by the National Institutes of Allergy and Infection, the National Institutes of Health, and the Diseases of Public Health Importance Training Grant.

Soh YQS, Moncla LH, Eguia, Bedford T, Bloom JD. 2019. Comprehensive mapping of adaptation of the avian polymerase protein PB2 to humans. eLIFE.  2019;8:e45079 DOI: 10.7554/eLife.45079