Quantifying influenza virus diversity and transmission in humans

We have previously shown that pandemic H1N1 and seasonal H3N2 viruses—both present during the first wave of the H1N1 pandemic in Hong Kong⁶—have similar transmission potential in household settings, and that antigenic variants of H3N2 co-circulated with clades of H1N1/2009.^6,7 In other parts of the world, and during the same time period, the unseasonal transmission of H3N2 was observed along with pandemic H1N1 virus.⁸ To characterize patterns of viral evolution at a finer-scale, and particularly the extent of virus genetic diversity that was transmitted among hosts, we performed whole genome deep sequencing on nasopharyngeal swabs collected from index cases with confirmed influenza along with their household contacts. Importantly, the household epidemiological information enabled us to assign donor/recipient pairs in suspected transmission events with relatively high confidence, compare these with unrelated pairs, and estimate spatio-temporal transmission chains.

The virus sample set was collected in July and August 2009 from 84 individuals (67 index patients and 17 other household members) living in Hong Kong; 16 patients were sampled twice, 2–4 days apart. We estimated intra-host virus diversity for each sample by mapping polymorphic sites onto the consensus genome assemblies to generate a list of single nucleotide variants (SNVs or minor variants) present at a frequency of at least 3%. Intra-host diversity was measured by the Shannon entropy, H, assuming site independence. Mean intra-host diversity was significantly higher (Wilcoxon rank-sum test p = 1.89e–12) for H3N2 (H = 33) than H1N1/2009 (H = 13). There was no significant Pearson correlation between high intra-host virus diversity and high viral titer⁷ (r = −0.3 for H1N1 and r = −0.16 for H3N2) for most of the genes, with the exception of PA and M for H1N1/2009 (Supplementary Table 1).

Phylogenetic analysis clustered whole genome consensus sequences by household for each group of patients diagnosed as infected with either H3N2 (Fig. 1) or H1N1/2009 (Supplementary Fig. 1). Comparisons of phylogenetic trees from each gene revealed no evidence for reassortment within this population during the time-frame of the study (data not shown). Three antigenic sublineages of H3N2 (A/Brisbane/10/2007-like, A/Victoria/208/2009-like, and A/Perth/16/2009-like) and three clades of H1N1/2009 (clades 3, 6 and 7) circulated in this population.⁶ Despite the relatively small population size, one case of mixed subtype infection was observed (patient 781_V1(0)), indicating that dual infection with seasonal and pandemic strains may not be a rare event.⁹

We compared SNVs across samples to determine if minor variants were shared within and between households. For both H3N2 (Fig. 2) and H1N1/2009 (Supplementary Fig. 2) we observed multiple positions in HA—including potential antigenic sites—where the minor variant nucleotide in one clade or lineage became the major nucleotide in another, with evidence of mixed infection at many other sites across the genome (Supplementary Figs. 3 and 4). For example, H3N2 households 707, 781, 671, 720 and 755 reveal a bimodal virus population that appears to have been transmitted intact in multiple transmission events. Overall, we tentatively estimated that approximately 66% of the H3N2-infected patients and 40% of the H1N1/2009-infected patients likely harbored mixed lineage infections (see Supplementary Table 2). To confirm these findings from the clinical specimens, we phased the SNVs into haplotypes by single molecule sequencing for 12 of the cell culture samples from 6 different households (Fig. 2 and Supplementary Fig. 2; Supplementary Tables 3–8). Notably, although the dominant haplotype indicates that each sample belongs to one major lineage, patients often carry a minor haplotype that resembles a separate lineage. This suggests that a number of the SNVs are not only de novo mutations that occurred in the index patient from a household, but are also shared across the community as a whole. We see a similar sharing of variant nucleotides when looking at global consensus sequences across seasons. Using HA consensus sequence data available in GenBank and human 2008 H3 sequences as a reference, we observed a shift of nucleotide frequency at some positions in subsequent seasons of H3N2 epidemics (Supplementary Fig. 5). This phenomenon is more pronounced for variants from the A/Victoria/208/2009-like lineage, in marked contrast to the decreasing trend observed for the A/Perth/16/2009-like lineage. However, no such trend was observed in pandemic H1N1 after the 2009 season. Additionally, frequency variations in H1N1/2009 are far less common than in H3N2. It is important to note that the A/Victoria/208/2009-like virus replaced the A/Perth/16/2009-like virus as the dominant lineage in recent years, leading in 2012 to a change of vaccine strain from A/Perth/16/2009-like virus to A/Victoria/361/2011-like virus (a phylogenetic subgroup of A/Victoria/208/2009). In contrast, pandemic H1N1 virus is antigenically stable and there was no change of vaccine strain after its introduction in humans in 2009. Overall, these data indicate that some viral lineages can be transmitted between individuals below current surveillance thresholds.

Since each virus sample collected will contain de novo mutations and potentially a mixed infection, we determined the similarity of the viral populations across the data set. To this end we calculated the genetic distance between samples by performing an all-versus-all pairwise comparison for each variant nucleotide position using an L1-norm (see Online Methods). We grouped pairwise comparisons by longitudinal pairs (same individual, sampled at two different visits), within households and across household pairs (Fig. 3). We determined that the median L1 genetic distances within household pairs or longitudinal pairs are significantly closer than any random pairing. This indicates that minor variants and their proportions can be used to infer inter-host transmission, even if a number of these correspond to co-infecting variants that are shared with individuals across households. Interestingly, for H1N1/2009 we see a number of “within household” pairs that are outliers (Fig. 3, dashed circle), providing further evidence of mixed infection. For example, variants present at a minor frequency in most of the samples from household 751 have become dominant in the visit 2 sample for the index case (751_V2(0)) (Supplementary Fig. 1). Although random sampling effects will impact mutational frequencies, such a profound increase in frequency is compatible with a selective advantage in that patient.

After excluding outliers and considering only a single sample (visit 1) per individual, there were 21 viable “within household” transmission pairs. To select other potential epidemic links within the community, we used the transmission and longitudinal pairs to identify outliers and determine a threshold of maximum genetic distance (after excluding outliers) (Fig. 3). Each pair was epidemiologically linked to a short transmission chain (see below). Using consensus sequences, we first inferred transmission networks across the population using a parsimony and graph-based algorithm.^10,11 We then used minor variant data to highlight potential localized outbreaks (Fig. 4) with cross-region links (i.e. Hong Kong Island, Kowloon and New Territories). This network agrees with the fact that there is a high volume of population flow within Hong Kong each day, allowing ample opportunity for influenza transmission across regions.

To further explore shared virus populations within households, we compared minor variants at each position in donor (index cases) and recipient transmission pair samples. Most variants found in the donor were shared with the potential recipient (Fig. 5, colored dots). The frequency of shared variants is much lower in pairs of unrelated samples (Fig. 5, black dots), although we find more shared variants in H3N2 than in H1N1/2009 pairs. We observed that the relative frequency of variants in the recipient is more often similar to that found in the donor, which is not the case for the same variants found in any other individual (Wilcoxon signed-rank test, p < 0.05), and implies the lack of a substantial genetic bottleneck at transmission. This in turn suggests that shared variants found in the recipient are not the result of de novo mutations but are more likely present in viruses that transmit between hosts and replicate.

From the household transmission pairs we estimated the probability that multiple variants are transmitted between hosts. In particular, polymorphic sites with variants only detected in the donor and those detected in both donor and recipient samples were selected to determine the probability of transmission as a function of variant frequency. Accordingly, for H1N1/2009, a donor variant found at a frequency of 10% has a 64% chance of being transmitted to the recipient; for H3N2, a donor variant at 10% has an 86% chance of transmission (Fig. 6). Because of the limited sample size it was not possible to determine with confidence the probability of transmission for variants present at frequencies below 10%.

To infer the size of the virus population before and after transmission that is able to generate productive progeny, we estimated the effective population size, N_e, by modifying a version of the Wright-Fisher (WF) idealized population model for our data. Specifically, for the donor/recipient pairs we took the frequency of the shared minor variants, p; the frequency of the major nucleotide at that position, q; and then calculated the variance of the difference in donor/recipient frequencies to obtain a variance effective size. For this we obtained a mean of 192 viral particles (median: 124; mean standard deviation (SD) range: 114–276) for H1N1/2009 and a mean of 248 (median: 138; mean SD range 47–457) for H3N2. To confirm the scale of our estimates, we utilized a different method based on the Kullback-Leibler divergence (KLD) (see Online Methods).¹² This gave a mean of 90 (median: 80; mean SD: 55) for H1N1/2009 and a mean of 114 (median 121; mean SD: 55) for H3N2. To estimate how many haplotypes would be present within these replicating populations, we used the phased SNV and reconstructed haplotype data and observed an average of 3 haplotypes for H1N1/2009 and 5 haplotypes for H3N2 transmitted across donor/recipient pairs (Supplementary Tables 3–8). The sample size is too small for the difference between H1N1/2009 and H3N2 to be significant. It is, however, theoretically possible that H3N2 has a higher N_e because the virus has been circulating in the human population since 1968 so that there is greater background genetic diversity and hence a greater diversity of lineages that can be transmitted among hosts. Crucially, these N_e and haplotype estimates suggest that multiple variants can be routinely transmitted between individuals, such that any transmission bottlenecks are fairly loose, and that a relatively small number of viral particles can initiate a productive infection with a number of variant strains that are co-transmitted.

In sum, we have analyzed minor variant dynamics in the transmission of influenza A virus within and across households during an epidemic and used that information to determine potential transmission events. The shared minor variant information between donors and recipients in transmission pairs was then utilized to estimate the number of viral particles that are able to infect and replicate in the recipient. The approach taken here could help define how prior immunity or other host factors, as well as virus subtype and strain, may affect transmission dose, of which our effective size estimates likely capture lower bounds. Indeed, this revealed the transmission of multiple variants, both from mixed infections and from within-host de novo haplotypes, indicating a relatively loose transmission bottleneck. Importantly, the shared variant data also suggest that there has been a single co-infection or super-infection event by two genetically distinct viruses during this epidemic, with this bimodal virus population then being transmitted intact in multiple subsequent transmission events. This is unsurprising in light of recent observations that natural selection can act on pools of virus variants linked by their co-localization in the same cell.¹³ In addition, this demonstrates that there are likely more cases of mixed lineages within infected patients than can be captured with standard consensus-based diagnostic assays. Such co-infections will obviously facilitate the occurrence of reassortment, and may help explain the frequent detection of reassortants between seasonal H3 viruses.¹⁴ Although similar observations have been made in animal studies,^11,15 this is the first demonstration for influenza A virus in humans. Characterizing the genetic information of transmitted virions allows a better understanding of influenza virus transmission in humans, and provides more accurate information for modeling epidemics and disease control strategies.