Fitness cost of reassortment in human influenza

Abstract

Reassortment, which is the exchange of genome sequence between viruses co-infecting a host cell, plays an important role in the evolution of segmented viruses. In the human influenza virus, reassortment happens most frequently between co-existing variants within the same lineage. This process breaks genetic linkage and fitness correlations between viral genome segments, but the resulting net effect on viral fitness has remained unclear. In this paper, we determine rate and average selective effect of reassortment processes in the human influenza lineage A/H3N2. For the surface proteins hemagglutinin and neuraminidase, reassortant variants with a mean distance of at least 3 nucleotides to their parent strains get established at a rate of about 10-2 in units of the neutral point mutation rate. Our inference is based on a new method to map reassortment events from joint genealogies of multiple genome segments, which is tested by extensive simulations. We show that intra-lineage reassortment processes are, on average, under substantial negative selection that increases in strength with increasing sequence distance between the parent strains. The deleterious effects of reassortment manifest themselves in two ways: there are fewer reassortment events than expected from a null model of neutral reassortment, and reassortant strains have fewer descendants than their non-reassortant counterparts. Our results suggest that influenza evolves under ubiquitous epistasis across proteins, which produces fitness barriers against reassortment even between co-circulating strains within one lineage.

Fig 1. Schematic of a reassortment process.

Two parent strains, p and p′, co-infect a host cell and produce a reassortant strain r. Here we focus on reassortment of the two surface proteins HA (blue segments) and NA (red segments); the reassortant strain r inherits one of these segments from each parent.

Fig 2. Tree representation of reassortment.

(a) Representation of reassortment in a two-segment genealogical tree. The parent strains p and p′ are in different sublineages of the tree; the reassortant strain r appears as a descendant of one of these parents (here p′; there is an equivalent tree in which r appears as a descendant of p). The strains p, p′, and r are the focal nodes of the clades C_p, C_p′, and C_r, respectively (grey areas). We identify the reassortment event by its set of core mutations, ${\mathcal{A}}_{p{p}^{\prime }}$, which appear on the segment that r inherits from p and generate the genetic distance between the parent strains in that segment. The core mutations appear on the branches between the nodes p and p′ (filled red triangles: mutations between p and the last common ancestor a, filled purple triangles: mutations between a and p′). Their reverse mutations appear on the branch between p′ and r (empty red and purple triangles), which can also contain additional mutations (grey triangles). (b) A true event (nr 1 in S2 Table) detected by our algorithm on the joint HA-NA tree. Each mutation on HA segment is labeled with a number between 1 and 1701 that indicates the site. The pattern of repeated and reversed mutations (filled and empty triangles) follows the scheme in Fig. 2a: the reassortant strain A/Hong Kong/CUHK33677/2004 is generated by an event with δ = 9 between p and p′ clades. (c) The result of a simulated reassortment event on the reconstructed genealogical tree, correctly detected by the algorithm. The internal node r is inferred as the reassortant ancestor of r_1/2,s, i.e. the strains evolved from the sequence that was actually generated by reassortment between p_s and $p_s^{\prime }$.

Fig 3. Fidelity of reassortment inference.

Histograms of reported HA-NA reassortment events for different core distances δ (red bars) are compared to expected number of false positives due to ambiguities in tree reconstruction, n₀(δ), from a null model of non-reassorting sequences (blue bars; error bars reflect the statistics over different realizations of the null model). The function n₀(δ) decreases exponentially with increasing δ (cf. Eq 6 in Materials and methods); the overall amplitude is set by the conservative assumption that all counts at δ = 1 are false positives. The resulting total number of false positives with δ ≥ 5 is below 1.

Fig 4. Reassortment of HA and NA in human influenza A/H3N2 from 2000 to 2015.

The 95 inferred events are mapped on a joint HA-NA tree. The reassortant strain r of each event is represented by a filled circle (color-coded by year of occurrence). The events are homogeneously distributed over the tree and the reassortant clades are predominantly at peripheral positions of the tree.

Fig 5. Negative selection on reassortment.

(a) The cumulative distribution of mean nucleotide distances d between reassortant and parent strains for the HA-NA reassortments in influenza A/H3N2 (red line) is compared to the corresponding distribution of distances for co-circulating strains in the same influenza season (solid blue line) and from the New York area only [28] (dashed blue line). (b) The ratio of reassortment counts to background counts in the interval d ≥ d_min (red circles) decreases with increasing lower threshold d_min and drops significantly below 1 (blue line). The suppression of reassortment at larger values of d signals distance-dependent negative selection. Bars show statistical errors due to the finite number of inferred reassortments. (c) The average number of strains in the reassortant clades with nucleotide distance ≤ τ from the focal node, 〈N_r〉(τ) (red line) is compared to the corresponding average number of strains in the parent clades, 〈N₀〉(τ). For τ ≲ 6, both functions increase with τ in an approximately exponential way; we estimate growth rates f_r ≈ 0.07 and f₀ ≈ 0.5, respectively (dashed lines; cf. Eq 1). The growth rate difference $\overline{s}\equiv f_0-f_r\approx 0.4$ measures the average fitness cost of reassortment. Bars represent statistical errors due to the finite number of counts (not shown when these errors are smaller than the dot size). See S3 Fig for an analogous inference based on amino acid distances.