The sequence of the HA receptor/fusion gene of the 1918 influenza A virus (Hsp) was determined in 1999 (ref. 2), followed shortly by the NA gene (Nsp)3, which codes for the viral sialidase. These molecules, which represent the major viral surface glycoproteins of influenza virus, are important targets of the host immune response, and critical changes in their structure can expand the viral host range and enhance the virulence of infection7,8. However, neither protein of the 1918 virus contains sequence motifs known to be associated with high virulence2,3. Attempts to account for the unusually high virulence of the virus have focused on the living conditions of soldiers, a highly susceptible group, and of civilians at the end of the First World War9, as well as on potentially unique properties of the virus itself10,11. The combination of Hsp and Nsp in an A/WSN/33 (WSN) (H1N1) genetic background was shown to be pathogenic in mice without prior adaptation of the proteins to growth in these hosts4,5, usually a prerequisite for producing a lethal infection in mice with a human virus. However, WSN itself is both adapted to and highly pathogenic in mice, so it was not possible to assess the possible contribution that Hsp and Nsp had made to the pathogenicity of the recombinant virus. Another candidate, the non-structural gene of the 1918 virus, had an attenuating effect when tested in a recombinant virus bearing the genetic background of A/Puerto Rico/8/34 (H1N1), also a highly pathogenic virus in mice10. Viruses generated with the matrix, nucleoprotein and non-structural protein genes of WSN replaced by the respective genes from the 1918 virus, together with Hsp and Nsp, were also pathogenic in mice4,5 but not significantly more than the recombinant possessing only the Hsp and Nsp genes. Thus, the molecular basis for the unprecedented virulence of the 1918 pandemic virus remains uncertain, opening the way for studies that directly test the contribution of Hsp and Nsp to pathogenicity.

We therefore used reverse genetics12 to produce a panel of recombinant viruses containing Hsp together with Nsp or the NA gene of A/Kawasaki/173/2001 (H1N1) (K173) in the background of the remaining negative-sense RNA gene segments derived from three influenza viruses: WSN, K173 and A/Memphis/8/88 (H3N2) (M88) (Table I). All recombinant viruses formed plaques on Madin–Darby canine kidney (MDCK) cell monolayers in the presence of trypsin and replicated efficiently to a high titre (data not shown).

Table 1 Properties of recombinant viruses

To assess the contribution of Hsp and Nsp to pathogenicity in vivo, mice were intranasally inoculated with the recombinant and parental viruses. Each virus possessing Hsp as well as the WSN virus replicated to a high titre in the lungs by day 3 post-infection (p.i.; Table I) and was pathogenic, resulting in serious morbidity and eventually death. Notably, the K173 and M88 parental human viruses did not cause discernible morbidity in mice and replicated to only low titres in the lungs, whereas the K173/Hsp and M88/Hsp viruses, in which the original HA proteins were replaced with Hsp, caused lethal infection in mice. The Nsp protein did not contribute to viral pathogenicity, as viruses containing Hsp and Nsp were not more pathogenic than those with the K173 NA. By contrast, the HA and NA proteins of WSN did not significantly enhance the virulence of either M88 or K173, as indicated by the lack of illness in mice inoculated with recombinant M88/HNWSN and limited illness associated with K173/HNWSN infection. The ability of WSN/HspNsp (Hsp and Nsp in the WSN background) and WSN to replicate in organs outside the respiratory tract was also assessed, with no evidence of replication detected in the heart, spleen, liver, kidneys or brain. Intracerebral inoculation of mice with 6 × 105 plaque-forming units (PFU) of WSN, WSN/HspNsp or K173/HspNsp (Hsp and Nsp in the K173 background) led to replication of only WSN in the brain (data not shown), an outcome consistent with the capacity of plasminogen binding by the NA to activate the HA and increase the neurotropism of WSN13, properties not shared by the NA protein of either the K173 or the 1918 virus. The relative pathogenicities of the viruses bearing Hsp depended on the type of virus providing the other gene segments. For WSN/HspNsp the dose required to kill 50% of infected mice (MLD50) was similar to that of WSN (Table I). MLD50 values for the other viruses were higher than for WSN/HspNsp, a reflection of the contribution of genes from the mouse-adapted WSN compared to the non-adapted human viruses. Thus, it appears that Hsp possesses unknown properties, not found in the WSN HA, that determine its ability to enhance the virulence of influenza A viruses in mice.

In a further effort to understand the increased virulence of the 1918 virus, we examined the lungs of mice intranasally infected with 10 MLD50 (see Methods for details of MLD50 calculation) of WSN, WSN/HspNsp, M88/HspNsp (Hsp and Nsp in the M88 background) or M88/Hsp, or a dose of M88 that was equivalent (in PFU) to 10 MLD50 of M88/HspNsp (as M88 was not lethal). By day 6 p.i. the viral antigen distribution for WSN/HspNsp, M88/HspNsp and M88/Hsp was considerably wider—encompassing the bronchus, bronchioli and almost the entire alveolar epithelium—than that seen for M88 or even WSN (Fig. 1a–c; see also Supplementary Figs 1 and 2; data for M88/HspNsp not shown). Infection with viruses possessing Hsp was characterized by massive recruitment of polymorphonuclear cells (mainly neutrophils) accompanied by intra-alveolar haemorrhage (Fig. 1d; see also Supplementary Figs 1 and 2). M88, on the other hand, produced limited inflammatory foci that resulted from initial infection of the bronchioli with only minor extensions into the adjacent alveoli. Most of the lung had a similar appearance to normal, uninfected controls (Fig. 1e, f; see also Supplementary Fig. 1) and only limited viral antigen was detected (Fig. 1; see also Supplementary Fig. 1). The presence of Hsp in M88/Hsp permitted greater viral replication over a longer period compared with M88, as shown by detection of viral antigen on day 6 p.i. (Fig. 1a, b), but the pathogenicity of viruses containing Hsp was not related simply to the extent of viral replication. High titres of virus were recovered from the lungs of mice infected with both WSN and WSN/HspNsp on both days 3 and 6 p.i., but lesions produced by WSN/HspNsp in lung tissue were clearly more severe (Supplementary Table 1 and Supplementary Fig. 2a, b). The lungs of mice infected with WSN had the characteristics of bronchopneumonia, but inflammatory reactions and viral antigens were confined mainly to the bronchus, bronchioli and peribronchiolar alveoli. As the M88/Hsp recombinant, which possesses only the HA from the 1918 virus, caused severe lung infection similar to that seen with WSN/HspNsp or M88/HspNsp infection, these results indicate that Hsp was the primary determinant of enhanced pathogenicity in the recombinant viruses. Severe lung infection was a hallmark of the illness produced by the original pandemic virus in humans6, suggesting a possible, although not conclusive, association between the pandemic virus HA and its pathogenicity in humans, and indicating the need to examine the contribution of the HA to viral pathogenicity in other animal models.

Figure 1: Pathological examination of lungs infected with the M88/Hsp or M88 viruses.
figure 1

a, Viral antigens were distributed throughout the entire lung of mice infected with M88/Hsp, including the bronchiole and alveoli of all lobes. b, By contrast, in lung tissue of mice infected with only the background virus M88, viral antigens were not detected on day 6 p.i. d, Global severe inflammatory reactions extending from the bronchus to the bronchiole to the alveoli were prominent in the lungs of mice infected with M88/Hsp. Numerous neutrophilic infiltrations were prominent in the peribronchial area and alveolar wall. Haemorrhage and oedematous lesions were common. These observations contrast with the lungs of mice infected with M88, which produced only small, irregularly distributed inflammatory lesions that were self-limited. Indeed, the vast majority of the lung was not affected (e), similar to normal lung (f). a, d, M88/Hsp, lung, 6 days p.i.; b, e, M88, lung, 6 days p.i.; c, f, control, lung. ac, Immunostaining with a rabbit polyclonal antibody to WSN; df, haematoxylin and eosin staining.

The severity of influenza-virus-induced disease in humans correlates with the ability of virulent viral strains to induce pro-inflammatory cytokines in macrophages14,15. Additionally, there is a strong association between cytokine levels and severity of clinical symptoms in people infected with influenza16,17,18. Therefore it is possible that pathogenicity of Hsp-expressing viruses might be causally related to their enhanced ability to induce pro-inflammatory cytokines and inflict immune damage to tissues in infected mice. To probe this possibility, we compared levels of cytokines/chemokines in lungs between viral isolates that differed in their virulence. Notably, the levels of monocyte chemotactic protein 1 (MCP-1), macrophage inflammatory protein (MIP)-1β, MIP-2, MIP-3α, interleukin (IL)-1β, IL-6, IL-12 (p40), IL-18 and granulocyte colony stimulating factor (G-CSF) in the lungs of M88/HSP-infected mice were substantially higher than in M88-infected mice (Fig. 3). Because activated macrophages are the major source of IL-1β, IL-6, IL-12, IL-18 and G-CSF during inflammation, the presence of high levels of these cytokines is strongly indicative of macrophage activation (Supplementary Fig. 3). Whereas the chemokines MCP-1, MIP-1β and MIP-3α are capable of inducing mixed leukocyte recruitment to the site of inflammation, the strong induction of MIP-2 (a strong chemoattractant for neutrophils) is highly consistent with the infiltration of granulocytes in the lungs of M88/HSP-infected mice. Taken together, these data suggested that Hsp is a critical determinant of macrophage activation, especially early on in infection (see day 1 data in Fig. 3), and of production of chemoattractant for neutrophils, which subsequently leads to the trafficking of neutrophils and acute lung injury19,20.

Figure 3: Cytokine expression in lungs of mice infected with M88/Hsp and M88 or mock infected (PBS).
figure 2

The expression of cytokines and chemokines in lung homogenates was determined by ELISA on days 1, 3, 5 and 6 p.i. Data are presented as the mass of cytokine (pg) per g of lung tissue (± standard deviation). (See also Supplementary Fig. 3.)

The 1918 virus is thought to have originated from an avian source2. Hence, the relative ability of the HA to recognize the preferred receptor species for human viruses—sialic acid bound to a neighbouring galactose by an α2,6 linkage (NeuAcα-2,6Gal) in contrast to the preferred avian virus receptor, NeuAcα-2,3Gal21—is thought to affect its ability to infect humans and cause disease. Using a competitive binding assay, we showed that the Hsp preferentially recognizes NeuAcα-2,6Gal (Fig. 2). Although humans possess both types of receptor species in respiratory epithelia and are susceptible to infection with viruses bearing HA proteins of either receptor specificity22, all of the human viruses studied thus far preferentially recognize NeuAcα-2,6Gal over NeuAcα-2,3Gal21. Thus, our data suggest that the virus that eventually became the 1918 pandemic strain must have circulated in humans, or possibly another adaptive host (such as pigs) with human-type receptors, for a sufficient time to develop a preference for the human sialic acid receptor. This model gains support from the recently solved crystal structure of Hsp (refs 23, 24), which predicts that the protein would preferentially bind to the human receptor23.

Figure 2: Competitive binding assay for haemagglutinin receptor specificity.
figure 3

ad, The relative binding affinities of HA in viruses WSN/HspNsp (a), A/dk/HK/836/80 (b), A/Kawasaki/173/01 (c) and A/New Caledonia/20/99 (d) were determined by competitive inhibition of HRP-conjugated fetuin binding by NeuAcα2,3Gal (open circles) or NeuAcα2,6Gal (filled squares) sialylglycopolymers.

An additional factor influencing the severity of an influenza pandemic or epidemic is the level of pre-existing immunity to the virus. To assess the extent to which the human population would be protected from the emergence of a 1918-like virus, we compared the neutralizing activities in a panel of human sera obtained in 1999 from patients of widely differing ages. The majority of the serum donors younger than 15 yr of age, most of whom had recently been infected with an H1N1 virus, showed strong neutralizing activity against a recent H1N1 test virus, but only limited activity against the WSN/HspNsp virus (Fig. 4; see also Supplementary Table 2). By contrast, serum samples from donors who were alive in 1918 showed markedly high activities against the WSN/HspNsp virus compared with those against recent H1N1 and H3N2 isolates. Donors born between 1934 and 1978 had only limited neutralizing activity against the WSN/HspNsp virus. Thus, a large segment of the population would be susceptible to an outbreak of a 1918-like influenza virus. The only group with significant natural protection would be survivors of the 1918 pandemic, who still express high levels of antibodies against an antigen to which they were exposed to over 80 yr ago, a phenomenon referred to as original antigenic sin25,26.

Figure 4: Neutralization activities in human sera against a virus possessing Hsp and Nsp.
figure 4

Neutralization titres of sera from a broad cross-section of donors were measured against WSN/HspNsp (filled diamonds) and a recent human H1N1 virus (open diamonds) in a plaque reduction assay. Donors younger than 14 yr were recently infected with an H1N1 virus, with the exception of two cases of H3N2 infection. The history of influenza infection in other donors was unknown. The reciprocal titre of serum required to neutralize at least 50% of each virus (50% neutralizing titre) is plotted against the date of birth of the patients.

Analysis of sequence information derived directly from viral genes in the preserved tissues of people infected during the 1918 pandemic offers unparalleled opportunity to assess the contributions of the HA, NA and other viral proteins to the extreme virulence of this unique pathogen. Using a reverse genetics technique to generate a panel of reassortants containing different combinations of the Hsp and Nsp genes in varied genetic backgrounds, we show that the HA protein (but not the NA protein) from the 1918 virus is essential for highly virulent infection in an animal model. However, the virulence of influenza virus is probably a polygenic trait, in that ultimately we expect other gene products, such as the NS1 protein27, to be implicated in the phenotype of the 1918 virus. Sequence analysis of the Hsp gene by others suggests that it originated in birds2 and therefore is likely to be contained in viruses still circulating in the avian pool, where viral proteins remain relatively unchanged. Once the properties of the Hsp gene that gave rise to its lethal infectivity are better understood, it should be possible to devise effective control measures and to improve global surveillance networks for influenza viruses that pose the greatest threat to humans as well as other animal species.



Full-length HA (GenBank accession number AF117241) and NA (GenBank AF250356) genes from the 1918 influenza A virus were constructed by using successive polymerase chain reaction amplifications with 70-mer oligonucleotides and cloned into a plasmid vector (pPolI) containing the RNA polymerase I promoter and terminator to generate pPolI-1918HA and pPol-1918NA. Recombinant viruses WSN/HspNsp, K173/HspNsp, K173Hsp, M88/HspNsp, M88/Hsp, K173/HNWSN, M88/HNWSN as well as parental viruses K173, M88 and WSN were generated by reverse genetics as previously described12. Viruses were passaged twice on MDCK cells, in the presence of 1.0 µg ml-1 TPCK-treated trypsin in minimal essential medium (MEM) supplemented with 0.3% bovine serum albumin and antibiotics. Influenza viruses A/New Caledonia/20/99 (H1N1) (NewCal), A/duck/Hong Kong/836/80 (H3N1) (dk/HK/80) and another recent clinical isolate, A/Kawasaki/233/2001 (H1N1) (K233), were similarly propagated in MDCK cells. All procedures related to viruses containing genes from the 1918 virus were performed in a Biosafety level 4 facility at the National Microbiology Laboratory, Health Canada, in Winnipeg, or in enhanced BSL3 at the University of Wisconsin-Madison.

In vitro analysis of viral replication

Semi-confluent MDCK cells were infected at a multiplicity of infection of 0.001 PFU per cell with each virus maintained under the same growth conditions used for the propagation of viruses. The supernatant was sampled at times 0, 15, 24, 39 and 48 h p.i., and titres were determined by standard plaque assay on confluent MDCK cells in the presence of 1.0 µg ml-1 TPCK-trypsin. The results of duplicate assays are reported.

HA receptor analysis by solid-phase binding assay

Viruses were allowed to attach overnight at 4 °C in wells of 96-well plates as previously described28. Polymer-bound conjugates containing the sialic acid N-acetylneuraminic acid linked to galactose through either an α-2,3 or α-2,6 bond (Neu5Acα2,3LacNAcb-pAP and Neu5Acα2,6LacNAcb-pAP) were prepared as previously described29. Serial dilutions of each sialylglycoconjugate were prepared in phosphate-buffered saline (PBS) containing 0.01% Tween-20 (TPBS), and 25 µl was added to the wells containing the immobilized viruses together with 25 µl of 47 nM horseradish peroxidase (HRP)-conjugated bovine fetuin in TPBS. After incubation at 4 °C for 2 h, the wells were extensively washed with PBS, and 100 µl per well of o-phenylenediamine (0.4 mg ml-1) in 0.02% H2O2, 50 mM citrate-phosphate buffer, pH 5.5 was added. After incubation at 25 °C for 15 min, the absorbance at 490 nm was read.

Viral pathogenicity and lethal dose determination

Isofluorane-anaesthetized 6-week-old female BALB/C mice were intranasally inoculated with tenfold serial dilutions (three mice per dilution) of viruses in 100 µl of PBS. The mice were monitored daily for 21 days for disease symptoms and survival. The dose required to kill 50% of mice (MLD50) was calculated by the method of ref. 30.

To determine the extent of viral replication in mice, we intranasally inoculated animals with 10 MLD50 of M88/HspNsp, M88/Hsp, K173/HspNsp, K173/Hsp, WSN or WSN/HspNsp, or with a dose of non-pathogenic virus, M88 or K173, equivalent in PFU to 10 MLD50 of M88/HspNsp and K173/HspNsp, respectively. On days 3 and 6 p.i., the lungs, heart, spleen, kidneys, liver and brain were collected from mice infected with WSN and WSN/HspNsp (n = 5). On day 3 p.i., the lungs of mice infected with the remaining viruses were collected. The organs were homogenized, and virus growth was determined in each tissue by plaque assay on MDCK cells.

To assess the ability of viruses possessing the 1918 virus HA and NA genes to replicate in the central nervous system, we intracranially inoculated 4-week-old BALB/C mice with 6 × 105 PFU of WSN, WSN/HspNsp or K173/HspNsp. Brains were collected on day 4 p.i. and homogenized, and virus replication was determined by plaque assay on MDCK cells.

For pathological analysis mice were infected with 10 MLD50 of the M88, M88/HspNsp, M88/Hsp, WSN and WSN/HspNsp viruses. After mice were killed on days 3 and 6 p.i., the lungs were removed and fixed in 10% phosphate-buffered formalin. The tissues were then dehydrated, embedded in paraffin, and cut into 5-µm-thick sections that were stained with standard haematoxylin and eosin. For viral antigen detection, sections were processed for immunostaining by the two-step dextran polymer method (DAKO), with a rabbit polyclonal antibody to WSN used as the primary antibody.

All animal experiments were performed under an approved animal use document and according to the guidelines of the Canadian Council on Animal Care.

Cytokine/chemokine analysis

Isofluorane-anaesthetized 6-week-old BALB/C mice were intranasally inoculated with 10 MLD50 of M88/Hsp or 1.5 × 106 PFU M88. On days 1, 3, 5 and 6 p.i., lungs were collected from three mice infected with each virus and homogenized in PBS. MCP-1, MIP-1α, MIP-1β, MIP-2, MIP-3α, IL-1α, IL-1β, IL-6, IL-10, IL-12 (p40), IL-12 (p70), IL-18, G-CSF, granulocyte-macrophage (GM)-CSF, tumour-necrosis factor (TNF)-α, IP-10, RANTES, vascular endothelial growth factor (VEGF), interferon (IFN)-α, and IFN-γ were detected by enzyme-linked immunosorbent assay (ELISA) (R&D Systems).

Human serum neutralization assays

Serum samples were collected between 1998–2001 from donors whose ages ranged from 15 months to 102 yr. The sera were treated with receptor-destroying enzyme (RDE) (Accurate Chemical and Scientific Corp.) to destroy inhibitors of influenza virus replication. After inactivation of the RDE by treatment at 56 °C for 1 h, WSN/HspNsp and K233 were each incubated with twofold serial dilutions of each serum at 37 °C for 1 h. Remaining infectivity was determined by titration of the samples in a plaque assay on MDCK cells.