Mutual information networks reveal evolutionary relationships within the influenza A virus polymerase

Abstract

The influenza A (IAV) RNA polymerase is an essential driver of IAV evolution. Mutations that the polymerase introduces into viral genome segments during replication are the ultimate source of genetic variation, including within the three subunits of the IAV polymerase (PB2, PB1, and PA). Evolutionary analysis of the IAV polymerase is complicated, because changes in mutation rate, replication speed, and drug resistance involve epistatic interactions among its subunits. In order to study the evolution of the human seasonal H3N2 polymerase since the 1968 pandemic, we identified pairwise evolutionary relationships among ∼7000 H3N2 polymerase sequences using mutual information (MI), which measures the information gained about the identity of one residue when a second residue is known. To account for uneven sampling of viral sequences over time, we developed a weighted MI metric (wMI) and demonstrate that wMI outperforms raw MI through simulations using a well-sampled SARS-CoV-2 dataset. We then constructed wMI networks of the H3N2 polymerase to extend the inherently pairwise wMI statistic to encompass relationships among larger groups of residues. We included HA in the wMI network to distinguish between functional wMI relationships within the polymerase and those potentially due to hitchhiking on antigenic changes in HA. The wMI networks reveal coevolutionary relationships among residues with roles in replication and encapsidation. Inclusion of HA highlighted polymerase-only subgraphs containing residues with roles in the enzymatic functions of the polymerase and host adaptability. This work provides insight into the factors that drive and constrain the rapid evolution of influenza viruses.

Figure 1.

Uneven sampling of H3N2 polymerase sequences over time influences Shannon entropy and mutual information. (A) The distribution of complete H3N2 polymerase sequences on GISAID per year between 1968 and 2015. (B) Upper panels, Sliding window analysis of mutual information (MI) for residue pairs PA-350/PB1-469 and PB2-590/PB1-709. Sliding windows were constructed with a width of 5-years and a slide-length of 1 year. Lower panels, Plots of the frequency of amino acids for each residue over the period 1968 – 2015.

Figure 2.

Re-weighting of amino acid frequencies improves MI estimates for unevenly sampled data. (A) The distribution of SARS-CoV-2 Spike RBD sequences generated by our laboratory per month between May 1st, 2021, and April 30th, 2021 from Washtenaw County, MI. The red line shows the number of confirmed COVID-19 cases in Washtenaw County, MI over the same time period. (B) Representative distribution of sampled Spike RBD sequences used to simulate the uneven sampling of H3N2 polymerase sequences (see Figure 1A). (C) The distribution of Spearman correlation coefficients between the MI from the original Spike dataset and the unweighted, equal-weighted, or incidence-weighted MI of 100 sampled datasets.

Figure 3.

Coevolving residues with PB2–627. (A) Residues that coevolve with PB2–627 shown highlighted on the replicating-encapsidating dimer conformation of the Influenza C polymerase (PDB ID 6XZR) (Carrique et al. 2020). Highlighted residues are in dark red (PB2) or dark grey (PA). (B) Domain organization of the IAV polymerase with coevolving residues indicated.

Figure 4.

wMI network of the H3N2 polymerase (PB2, PB1, PA). Nodes represent residues and edges represent the normalized wMI (z-score) between residues. Residue nodes are colored red for PB2, blue for PB1, and grey for PA. An edge threshold was set at the normalized wMI score (58) that minimizes relative maximum subgraph size (see Figure S3). The network visualization was created using the associationsubgraphs package for R (Strayer et al. 2023).

Figure 5.

Features of the residues in Subgraphs 1 and 2 from the wMI network of the H3N2 polymerase. (A) Amino acid frequencies from 1968 – 2015 for the residues within Subgraph 1 (left) or Subgraph 2 (right). (B, C) location of the residues in Subgraph 1 (B) and Subgraph 2 (C) plotted on the replicating-encapsidating dimer conformation of the Influenza C polymerase (PDB ID: 6XZR) (Carrique et al. 2020). In B-C, highlighted residues are in dark red (PB2) or dark grey (PA).

Figure 6.

wMI network of the H3N2 polymerase (PB2, PB1, PA) and HA. Nodes represent residues and edges represent the normalized wMI (z-score) between residues. Residue nodes are colored as in Figure 4, plus orange for HA. HA residues that are located in antigenic regions A-E are shown in bold. Residue −6 (SS) is in the cleaved N-terminal signal sequence of HA. An edge threshold was set at the normalized wMI score (40.506) that minimizes relative maximum subgraph size (see Figure S5). The network visualization was created using the associationsubgraphs package for R (Strayer et al. 2023).

Figure 7.

Features of the residues in Subgraphs 4, 8, and 10 from the wMI network of the H3N2 polymerase and HA. (A) Amino acid frequencies from 1968 – 2015 for the residues within Subgraph 4 (top left), Subgraph 8 (bottom left), or Subgraph 10 (right). (B-D) location of the residues in Subgraph 4 (B), Subgraph 8 (C), or Subgraph 10 (D) plotted on the post-cap-snatching conformation of the H3N2 polymerase (PDB ID: 6RR7) (Fan et al. 2019). In B-D, highlighted residues are in dark red (PB2), dark blue (PB1), or dark grey (PA).