The 1957 and 1968 influenza pandemics, each of which killed ≈1 million persons, arose through reassortment events. Influenza virus in humans and domestic animals could reassort and cause another pandemic. To identify geographic areas where agricultural production systems are conducive to reassortment, we fitted multivariate regression models to surveillance data on influenza A virus subtype H5N1 among poultry in China and Egypt and subtype H3N2 among humans. We then applied the models across Asia and Egypt to predict where subtype H3N2 from humans and subtype H5N1 from birds overlap; this overlap serves as a proxy for co-infection and in vivo reassortment. For Asia, we refined the prioritization by identifying areas that also have high swine density. Potential geographic foci of reassortment include the northern plains of India, coastal and central provinces of China, the western Korean Peninsula and southwestern Japan in Asia, and the Nile Delta in Egypt.
Simultaneous infection with multiple influenza virus strains can affect virus fitness components, such as virus growth performance, and thus affect virus pathogenicity, transmission, or recombination (1). In a host infected with 2 closely related influenza viruses, the stains can reassort, exchanging gene segments to produce new strains, some of which might have increased virulence. Virulence might also trade off with transmission such that more pathogenic viruses spread more slowly (2). However, in some instances, a reassortant virus can have high transmissibility and high pathogenicity. For example, reassortment between influenza viruses of humans and birds resulted in the 1957 and 1968 pandemic viruses, each of which is estimated to have killed ≈1 million persons (3,4). The exchange of genes between pairs of influenza virus subtypes increased virulence in animal models, including reassortment between subtypes H9N2 and H1N1, between H5N1 and H1N1, and between H3N2 and H5N1 (5,6). We focus on reassortment between subtypes H3N2 and H5N1 because extensive data are available, but given sufficient data, our approach could be extended to other subtypes.
A caveat is that even if virus subtypes H3N2 and H5N1 were to reassort in swine, the spread of the reassortant virus among humans might require further virus adaptation events; for example, mutations might be required for the virus to replicate efficiently in humans or to be transmitted among humans (22). Recent work has shown that as few as 5 aa substitutions are required for aerosol spread of subtype H5N1 among mammals (23). With these qualifications in mind, this analysis provides actionable recommendations about which areas to target for intensified farm and market surveillance. Such surveillance could enable early detection of a reassortant influenza (H3N2/H5N1) virus, should it arise in swine, and facilitate containment of the virus before it crosses the species barrier to humans.
Our finding that in China the probability of subtype H3N2 infection increases with human population density is compatible with previous studies that detected a positive association between population, influenza cases, and mortality rates (12,24). Reasons for this association could be that the number of susceptible human hosts increases with population (11) or that surveillance efforts are greater in populous areas (25). Our results with regard to subtype H5N1 in birds are also largely consistent with those of previous studies that mapped subtype H5N1 hotspots in China and Egypt. In China, several provinces identified as having high ecologic suitability for subtype H5N1 (including Shandong, Jiangsu, and Sichuan) were also identified as subtype H5N1 hotspots in a previous study that used a different statistical model and different predictor variables (11). In China, previous analyses have concluded that risk for subtype H5N1 increases with the density of domestic ducks (26). In Egypt, earlier studies identified high-intensity crop production as a statistically significant predictor of subtype H5N1 in poultry (27). Similarly, we found that subtype H5N1 infections in poultry were associated with poultry density, which was highly correlated with crop production. In a previous study, models constructed from satellite images of vegetation predicted that the highest environmental suitability for subtype H5N1 is along the Nile River and in the Nile Delta (28). Our models were constructed from different predictor variables, such as poultry density, but yielded similar results: the highest number of subtype H5N1 cases in poultry were predicted to occur in districts in the Nile Delta.
Efforts to contain the A(H1N1)pdm09 virus would have been more effective if the virus had been detected in animal populations before it was transmitted to humans (29). Continuous zoonotic influenza surveillance is needed in China and Egypt and requires a network of laboratories to screen surveillance samples and requires financial incentives to encourage poultry producers and sellers to report outbreaks. One strategy for early detection of a reassortant virus could involve increasing farm and market surveillance in the identified areas (i.e., live-bird markets in 6 provinces in China [Guangdong, Hunan, Jiangsu, Shanghai, Sichuan, and Zhejiang] that have a >50% chance of subtype H3N2 and H5N1 co-occurrence and above average swine density). Increased monitoring could identify hotspots where subtype H5N1 is circulating, leading to more efficient targeted vaccination of poultry, and could pinpoint prefectures at high risk for a reassortant virus. In China, sanitary practices, such as cage disinfection and manure disposal, would substantially reduce risk for subtype H5N1 in live-bird markets (30).
In Egypt, our results support increased surveillance of backyard flocks near Benha, Cairo, Dumyat, El Faiyum, Shibin el Kom, and Tanta, where suitability for subtypes H5N1 and H3N2 is predicted to be high. Control measures could include compensation plans and vaccination of poultry with a recently developed subtype H5N1 vaccine that is more effective than previous vaccines against strains circulating in Egypt (10). Reporting of poultry disease outbreaks in Lower Egypt is poor (31), probably because farmers fear loss of income if authorities cull their flocks. Indeed, birds suspected to be infected with subtype H5N1 are often sold quickly at a discount, resulting in virus transmission to buyers’ flocks and families (32). If equitable compensation schemes were implemented, reporting of subtype H5N1 might increase and outbreaks could be contained more quickly, reducing opportunities for subtypes H5N1 and H3N2 to co-infect humans or domestic animals and, thus, for reassortment.
In general, policies such as culling must have a scientific basis because these measures have major effects on the economy and animal welfare. For example, when part of a swine herd is culled to contain an outbreak, it might become necessary to euthanize the entire herd, including animals with no influenza exposure, because buyers will not accept them (33). Furthermore, influenza outbreaks among livestock can trigger major global declines in meat prices, and the nature and timing of veterinary health authorities’ responses to an outbreak can affect the extent to which demand recovers after the crisis. In particular, when control measures such as culling are scientifically well justified and explained to the public soon after the start of an outbreak, consumer confidence is restored more quickly (34).
Although our maps suggest a risk for reassortment in Lower Egypt and eastern and central China, in vivo reassortment of subtypes H3N2 and H5N1 has not been detected in humans in these areas. On the other hand, numerous infections with influenza (H3N2)v, a reassortant virus that contains genes from a subtype H3N2 virus circulating in swine and from the A(H1N1)pdm09 virus, have been detected in humans in North America (35,36). This finding raises the question of why subtype H3N2v has spread but subtype H3N2/H5N1 reassortants have not. Spread of subtype H3N2v could result from the fact that the reassortant virus contains the M gene from the A(H1N1)pdm09 virus, which increases aerosol transmission (35,37). Our models might explain why, in contrast with subtype H3N2v reassortants, no subtype H3N2/H5N1 reassortants have been detected in humans. For example, we predict that subtypes H3N2 and H5N1 occur in Hunan, China, a province that has high swine density and was the geographic origin of subtype H5N1 viruses in clade 2.1 (38). Influenza (H3N2/H5N1) reassortants in which the nonstructural gene comes from a clade 2.1 virus replicate poorly in mice (5). Thus, subtype H3N2/H5N1 reassortants might not have emerged as often as subtype H3N2v reassortants because the provinces where subtypes H3N2 and H5N1 overlap contain a clade of subtype H5N1, whose genes reduce the fitness of reassortant viruses. If this hypothesis is correct, if subtypes H5N1 and H3N2 infect a pig in central China and exchange genes, the hybrid virus might not replicate efficiently or transmit to other hosts. Furthermore, a reassortant virus with surface proteins similar to those of subtype H3N2 viruses that have circulated in humans recently might have poor transmissibility because of preexisting immunity (18).
Applying our modeling framework to other zoonotic influenza subtypes, such as H3N2v, could yield insight about geographic hotspots of reassortment and the pattern of spatial spread of reassortants. To accomplish this, 2 data limitations must be overcome. First, to be incorporated into spatial models, influenza sequences submitted to GenBank or GISAID should be accompanied by geographic data at relatively high spatial resolution, for example, names of cities or counties where sampling was conducted. However, such sequences are often accompanied by only the state or country of the sample, which reduces the usefulness of the data for fine-scale spatial modeling (39). For example, we searched online databases and confirmed that the geographic data available for Indonesia are insufficient to construct a spatial model to predict sites with a high risk for reassortment. Second, more extensive surveillance of livestock is needed to provide sufficient sample sizes to parameterize geographic models. Currently, the number of influenza subtype H1, H3, and H5 viruses from swine in major databases is an order of magnitude lower than that available for humans (Technical Appendix Table 3 [PDF - 1.61 MB - 9 pages]). Additional surveillance of swine could lead to better predictions about hotspots of influenza in livestock and sites of potential swine-to-human transmission. Livestock surveillance campaigns should sample large geographic areas and include regions where production is high (35).
The potential for reassortment between human and avian influenza viruses underscores the value of a One Health approach that recognizes that emerging diseases arise at the convergence of the human and animal domains (29,40). Although our analysis focused on the influenza virus, our modeling framework can be generalized to characterize other potential emerging infectious diseases at the human–animal interface.