Assessment of baseline age-specific antibody prevalence and incidence of infection to novel influenza A H1N1 2009

Virus isolation and culture of H1N1 2009

Immediately after the notification of a widespread H1N1 2009 outbreak in the USA and Mexico, the Respiratory Virus Unit (RVU) of the HPA identified A/Aragon/R3218/08(H1N1) as a suitable substitute virus for initial assay set-up and validation purposes both for molecular and serological work. The virus, which was classified as a BSL2 + pathogen, was received at the end of April and cultivated in eggs to provide material for positive controls in real-time polymerase chain reaction (RT-PCR) development and also as antigen for the initial stage of serological assay development. The HPA Centre for Emergency Preparedness and Response (CEPR) also received this virus to generate standard ferret sera.

Upon identification of the first clinical cases in the UK (molecularly confirmed by HPA RVU) virus isolation work involving cell culture at BSL3 started from virus-positive clinical swabs taken from these cases. Once isolates from the first UK cases were grown and genetically characterised, a suitable representative was provided to the National Institute for Biological Standards and Control (NIBSC) to attempt generation of an rg virus and preparation of ferret serum standards. NIBSC was also tasked by the WHO with the creation of an rg virus for the A/California/7/09 prototype US strain, which it successfully generated at the end of May 2009 (NIBRG121). The rg version of the first UK isolate (A/England/195/09) became available from NIBSC in the second half of June 2009 (NIBRG122). It was necessary to compare the properties of these and other available viruses to determine the optimum virus to use for large-scale serology studies, a process which is important to ensure that the choice of strain for extensive serology is representative of circulating strains.

Generation of animal sera/antigenic analysis

There is at least a 3-week time delay between the availability of a particular virus strain and availability of a specific ferret standard post-infection serum. A minimum time is required following experimental inoculation of ferrets to develop high-titred antibody. Similar to the strategy for virus cultivation, generation of antisera was performed in stages and a ferret post-infection serum to A/Aragon/R3218/08 became available first (end of May), followed by ferret serum to A/California/7/09 and A/England/195/09. These serum standards and virus prototype strains were used to establish the antigenic relatedness of the various H1N1 2009 isolates that were available at that time from Europe and the USA. In this way, a robust strain selection process was undertaken to ensure optimisation of serological assays used by HPA for seroepidemiological and clinical diagnostic purposes, as well as for the analysis of pandemic vaccine trials.

Strain selection

Table 1 indicates the reactivity of various virus isolates with panels of reference ferret antisera in HI tests. The highest reactivity (titre) is expected in reactions between virus strains and post-infection antisera from ferrets inoculated with the same (homologous) virus. It can be seen that the antigenic reactivity of NIBRG121 and NIBRG122 rg viruses is identical to their respective prototype or parental wild-type strains using a panel of post-infection ferret antisera raised against pandemic, seasonal and older H1N1 strains. This also demonstrates the antigenic relatedness of prototype UK strain to the A/California/4/09 prototype strain and its rg equivalent virus (NIBRG121). Together, these data validate the choice of an rg virus as a suitable alternative to wild-type strains and indicate that sera related to the outbreak in the UK could be analysed with the NIBRG122 virus.

HI protocol development

Influenza viruses show variation in their capacity to bind RBCs from different species. The ability of HA molecules to bind surface receptors on the RBC depends on which type of sialic acids these receptors contain and what linkage connects the glycan residues. Human viruses agglutinate RBC from chicken/turkey, human and guinea pig, but not from horse. In contrast, avian viruses agglutinate RBC from all four species, whereas the binding of swine viruses depends on whether they have human or avian-like HAs. Assessment of RBC binding of virus strains is a surrogate measure of receptor specificity and knowledge of this property is essential for interpretation of serology, as incorrect use of RBC indicator for the HI bioassay will lead to an underestimate of available antibody, as has been observed in measuring antibody responses to avian virus infection and H5 vaccine responses.

To identify an appropriate source of RBC for the HI assay, empirical investigation of animal RBCs from different species was undertaken to optimise for the A/H1N1 2009 virus. RBCs from chicken, turkey, guinea pig, pig, horse and human (type O) were investigated for adjustment of the HI assay. Turkey RBCs provided the highest sensitivity (Table 2).

Serum validation panels

After development of suitable bioassay protocols, the performance of HI and MN assays was evaluated using archived serum panels. These sera were collected prior to 2009 (i.e. from a population without exposure to swine influenza viruses) and were used to serve in establishing the levels of cross-reactive antibodies in various groups of the population. The panels are described as follows:

1. England age-stratified panels (collected in June 2003 and June 2004) Samples were geographically representative of England and were obtained from residual serum samples from outpatient visits.

2. Influenza-like illness panel (2002/03) These were paired sera from UK patients with ILI, many of whom had serologically confirmed seasonal influenza.

3. Influenza-like illness panel (SARS 2003) These were UK sera collected in 2003 during the SARS (severe acute respiratory illness) incident from patients with suspected SARS infection for which recent seasonal influenza had been excluded.

4. Seasonal Vaccine Study Trivalent inactivated vaccine trial in healthy adult and elderly volunteers (trial performed in southern hemisphere in 2005).

Using the NIBRG122 strain, significant levels of cross-reactive antibody were found in sera without primary exposure to H1N1 2009 virus, using both HI and MN (although analysis was performed with both assays and yielded similar results, we discuss HI only in this section for simplicity). Results of the HI analysis of these four panels are summarised in Table 3. The detection limit for HI is a titre of 1 : 8 (for MN it is 1 : 10).

For the England age-stratified panel, a total of 616 samples collected in July 2004 from four different regions in England were selected to determine the specificity of the HI test. The age range was 0–79 years and 49% were male. Table 3 shows the number of samples by age group, proportions of individuals with HI titres above the cut-off points ≥ 1 : 8 and ≥ 1 : 32, as well as geometric mean titres (GMTs) by age group (information about region was neglected, as regions were not evenly represented in this collection).

We found evidence for cross-reactive antibody in all, except the 1- to 4-year age group. Our analysis suggested that the observed cross-reactivity is age specific, with children under the age of 10 years having little or no cross-reactive antibody, whereas in older adults over the age of 65 years more than 20% of all tested individuals showed titres above the detection limit.

We also observed cross-reactive antibody when analysing sera from subjects with a clinical diagnosis of ILI and severe respiratory illness (SRI) or volunteers from a vaccine trial of seasonal vaccine (which were collected during the 2002–3 season) in 2003 and 2005, respectively. Samples in these panels were grouped into two major age groups (under and above 55 years of age) and, where possible, were analysed as acute/convalescent or pre-/post-vaccination serum pair.

Sample size

The study was designed to provide an estimate of the proportion of the population in defined age groups with antibodies against H1N1 2009 before and after the pandemic, and at monthly intervals during the pandemic. The monthly numbers were constrained by the need to generate data quickly and inevitable limitations on laboratory capacity. The aim was to obtain samples spread evenly by age groups (< 5, 5–14, 15–24, 25–44, 45–64, 65–74, 75–79, 80 + years) and regions in the baseline, with each age group having about 200 samples at baseline and post pandemic. The age groups were chosen to match those in the Royal College of General Practitioners (RCGP) influenza surveillance data set with expansion of the 65-and-over age group in order to define with greater precision the relationship between birth cohort and prevalence of pre-existing cross-reactive antibodies to H1N1 2009. A sample size of 1600 (i.e. 200 in each age group) would provide 95% CIs, which were deemed reasonably narrow; for example, for a prevalence of 5%, 95% CI would be 2.9 to 9.0, and for a prevalence of 25% the corresponding 95% CI would be 19.2 to 31.6. This was deemed to provide sufficient precision for meaningful estimation of baseline seroprevalence.

For the monthly incidence estimates, for which there was a testing limit of about 1000 per month, all of those aged 65 + years were collapsed into a single age group. A sample size of 200 per age group would allow prevalence to be estimated with 95% CIs ranging from ± 5% around a prevalence of 10% to ± 7% around a prevalence of 50%. The 95% CIs around various overall incidence estimates based on 1000 samples at each time point are shown in Table 4. Within each age group the precision of the incidence estimate would be much lower.

Analysis of baseline seroepidemiology

To investigate whether certain birth cohorts had higher baseline immunity geometric mean HI and MN titres with 95% CIs were calculated by 10-year birth cohorts, as well as the proportions with titres greater or equal to the detection limit and four times the detection limit (i.e. ≥ 1 : 8 and ≥ 1 : 32 for HI and ≥ 1 : 10 and ≥ 1 : 40 for MN) with exact binomial 95% CIs. Results from the HI assay were semiquantitative (in powers of two); the highest titre to which dilutions were made was 1 : 1024. For the MN assay, results were continuous, with an upper dilution of 1 : 320 (a small number of samples were diluted beyond this titre). A multivariable normal errors regression model was also fitted to log(HI) and log(MN) results to investigate the effects of birth cohort/age, sex and region. For all statistical analyses, titres for samples that fell below the detection limit were set to one-half of the detection limit (i.e. four for HI and five for MN), and titres greater or equal to the highest detection limit were set to the maximum detection limit.

Analysis of seroprevalence over time

Seroprevalence over time, based on an HI titre of ≥ 1 : 32, was investigated using all samples taken from August 2009 to April 2010. A logistic regression model was fitted with factors for month, gender, age, region and laboratory type (microbiology or chemical pathology). Interactions were also investigated. This modelling approach was then used to obtain predicted probabilities of a seropositive result by month, age and region (London/outside London).

Estimation of seroincidence from observed changes in seroprevalence

To investigate seroincidence over the course of the pandemic, seroprevalence at three time points [pre-pandemic (2008), post-summer wave and post-autumn wave] was estimated and the change from one time point to another was assumed to reflect seroincidence. The post-summer wave period (weeks 34–37 2009) and autumn wave period (January to April 2010) were defined based on the HPA-estimated case numbers by week and allowing time for seroconversion by HI (at least 3 weeks). An initial descriptive analysis was performed using age-specific reverse cumulative distribution curves of HI titres at the three time points. Age-specific seroincidence was then estimated from baseline to post-summer wave, post-summer wave to post-autumn wave, and baseline to post-autumn wave by calculating the difference in seroprevalence at these time points with 95% CIs. This was carried out using Wilson’s score method, implemented in the Epi library in the r software (The R Foundation for Statistical Computing, Vienna, Austria). Because of the different age composition of the collections, direct age standardisation within the 65-and-over age group, using smaller age groups of 65–74, 75–79 and ≥ 80 years, was required.

Due to time constraints in the laboratory, not all samples could be tested using both HI and MN. Thus, seroincidence was measured using changes in HI from baseline to post-summer wave, and baseline and post-summer wave to post-autumn wave, and by HI and MN from baseline to post-autumn wave. Again due to limited time for testing, the only samples tested that were collected in November and December were from individuals aged under 45 years, in whom changes in incidence were likely to be greatest. Further, some samples could not be tested with MN as there was not enough serum left.

Seroincidence was estimated separately for samples collected in London and samples collected elsewhere. The reason for stratifying by London and elsewhere was that London was one of the regions that experienced a substantial H1N1 2009 wave during the summer. In addition, the demographic structure of the population in London is different to that in other regions, with a higher proportion of individuals in the younger age groups, which, together with the higher population density, may result in a higher R-value for London than elsewhere. As not all regions contributed samples at all time points, a sensitivity analysis was carried out for which seroincidence was estimated for only those regions with samples available from both 2008 (baseline) and 2010 (post-autumn wave).

All statistical analyses were carried out in r version 2.11.1.

Likelihood-based estimation of seroincidence

Estimating the incidence of infection by measuring monthly changes in antibody prevalence from the pre-pandemic baseline has certain limitations. First, the monthly samples are distributed over a 30-day period, during which incidence may be changing, particularly at the height of the wave. Second, the variable time to seroconversion between individuals means that even if all samples were taken on the same date they reflect incidence at different times in the previous weeks. Third, derivation of incidence by comparing prevalence between time points reduces the precision of the estimate for a given sample size and may result in negative point estimates for groups in which incidence is low.

A novel method based on a likelihood approach was therefore developed to overcome some of the limitations of the more conventional statistical method described above. The new method uses information on the distribution of the seroconversion intervals in confirmed cases and the exact date of each serum sample, and allows integration of information on the temporal distribution of cases from clinical surveillance (as coming, for example, from the HPA clinical case estimates2). To link information on distributions of cases from surveillance with the sequential serological samples, we proceeded in three steps:

1. For each age group, we constructed a set of putative H1N1 incidence curves (i.e. number of new infections people per week per age group) by assuming that the proportion of people seeking health care was constant during the epidemic. We thus obtained a set of age-group-specific putative incidence curves, built by multiplying the proportion of cases from surveillance in the considered age group in each week by a constant (as the weekly proportion of cases sums to one, this constant is also the final cumulative incidence in the age group). The inference algorithm aims to assess one parameter per age group.

2. For each of the parameterised incidence curves, using the distribution of intervals from onset of symptoms to seroconversion, we redistributed cases to obtain a seroconversion curve, describing how many people seroconvert daily in each age group. By integrating these cases over time starting from the baseline (2008), for each age group, in turn, we linked the incidence curves with the expected change in antibody prevalence.

3. With the daily value for the antibody prevalence given the final cumulative incidence, we could finally derive a likelihood function for the sequence of serological samples. For this, we assumed that the probability of picking a seropositive sample followed a binomial distribution. We inferred the most likely incidence curve by using maximum-likelihood methods.

To describe the kinetics of seroconversion after infection in step 2, we fitted a scaled Weibull distribution to the subset of samples from confirmed cases (n = 115) with a known date of onset of symptoms. Taking as definition of seroconversion an achievement of a HI titre ≥ 1 : 32, it was possible to derive a likelihood function for a given set of parameters of the scaled Weibull distribution (scaling constant in addition to the two parameters of the Weibull distribution). Assuming that the seroconversion after symptoms follows a Weibull distribution, the probability of observing a positive sample after an interval of n days is equal to the probability of the cumulative Weibull distribution. We derived the parameters of the interval to seroconversion distribution using a maximum-likelihood algorithm.

This method does not require a pre-pandemic baseline (although it uses the information if available) and by using information on date of sample it allows the generation of a continuous curve, describing how incidence is changing over the course of the pandemic. It also removes the possibility of generating negative incidence estimates by sampling error.

This likelihood-based estimation method was applied to the serological data in the age groups (1–4, 5–14, 15–24 and 25–44 years) for which there was a continuous supply of samples throughout the two waves of the pandemic for both London and non-London regions. The temporal distributions of cases by week in the first and second waves based on health-care consultations in London and non-London regions were used to generate potential incidence curves by multiplying them by putative total incidences for the two waves. The most likely incidence given the observed temporal sequence of serological samples with HI titres ≥ 1 : 32 in the given region is then estimated. The CIs were estimated using 10,000 bootstrap samples drawn from the incidence distributions of the different age groups. The baseline seroprevalence was assumed to be the same in and outside London.

Pre-pandemic HI/MN correlation

Figure 10 shows MN titres plotted against HI titres with a loess smoother for the 1169 samples tested with both assays. Spearman’s rank correlation r between the HI and MN titres was 0.68. There is therefore some evidence of positive correlation between the assays. Of the 823 samples with HI < 1 : 8, 162 (19.7%) had MN titres of ≥ 1 : 10 and 42 (5.1%) had MN titres of ≥ 1 : 40.

FIGURE 10.

Haemagglutination inhibition vs MN titres, all age groups, for samples collected in 2008. The dashed lines show the minimum detection limit for the two assays (1 : 8 and 1 : 10 for HI and MN, respectively). The solid line is a loess smoother.

Correlation between MN and HI titres in seroincidence samples

Figure 11 shows MN titres plotted against HI titres for the 1148 samples collected between 1 January and 19 April 2010 that had matched valid results on both MN and HI. Note that the highest dilution titrated to was an MN value of 320 so results are censored at this titre.

FIGURE 11.

Haemagglutination inhibition vs MN titres, all age groups, for samples collected January to April 2010. The dashed lines show the minimum detection limit for the two assays (1 : 8 and 1 : 10 for HI and MN, respectively). The solid line is a loess smoother.

Spearman’s rank correlation r between the HI and MN titres was 0.79. As seen with the baseline samples, a large proportion of those with an HI titre of < 1 : 8 had MN ≥ 1 : 10 (136 out of 1148 samples, 11.8%).

Figure 12 shows the proportion of discordant samples by age group, i.e. the number of samples with detectable antibody on MN but not on HI [MN positive (MN +)/HI negative (HI–) as a proportion of all samples in that age group, and vice versa] in the baseline (Figure 12a) compared with the post-pandemic sera (Figure 12b). For the baseline sera the proportion of samples MN +/HI– increases with age (chi-squared test for trend p < 0.001) but not for samples HI +/MN– (chi-squared test for trend p = 0.30). Similar results were obtained for the post-pandemic sera, with the proportion of samples MN +/HI– increasing with age (chi-squared test for trend p < 0.001) but not for samples not for samples MN–/HI + (chi-squared test for trend p = 0.94.)

FIGURE 12.

(a) Baseline (2008). (b) Post second wave (January to April 2010).

Among younger age groups (< 25 years) the point estimate of the proportion of discrepant samples with detectable antibody by MN but not by HI was smaller in the post-pandemic sera than in the baseline sera.

Baseline antibody prevalence

The results of the baseline prevalence survey showed that a substantial proportion of older adults had pre-existing cross-reactive functional antibodies capable of neutralising H1N1 2009, presumably as a result of prior exposure to antigenically related H1N1 influenza viruses circulating in previous decades, or as a result of heterosubtypic antibodies capable of providing broad cross-subtype protection. Examples of antibody that have broad neutralising capability across multiple influenza A subtypes have recently been described. ^11–13 During every winter epidemic of influenza, it is estimated that between 10% and 20% of the population are infected with circulating influenza viruses. Virological surveillance in the UK since the 1950s has confirmed the presence of circulating influenza viruses during winter epidemics of ILI. In the twentieth century, the circulating viruses have been influenza A viruses H3N2, H2N2 and H1N1, at various times, or indeed co-circulating. The older an individual, the more influenza seasons and natural infections he/she will have experienced. Increasing time of exposure to influenza A viruses, including H3N2, H2N2 and H1N1, will increase the likelihood of an individual having a repertoire of heterosubtypic antibodies to conserved epitopes of viral proteins. The presence of such antibodies, irrespective of their derivation, is predictive of the probability of protection from infection with H1N1 2009, and is consistent with the lack of observed impact of H1N1 2009 in older age groups.

The prevalence of pre-existing antibodies increased significantly in those born before the 1950s with a further substantial rise in those born before the 1930s. Viruses of the A/H1N1 subtype circulated in humans from 1918 until they were replaced by A/H2N2 influenza, which caused the pandemic of 1957. During this period of circulation, A/H1N1 viruses underwent significant antigenic drift away from their 1918 virus progenitor. 14 In 1977, A/H1N1 influenza viruses from the early 1950s re-emerged in humans, and, until 2009, went through further substantial antigenic evolution in the human host. 15 By contrast, classical H1N1 influenza viruses in swine remained antigenically relatively static until 1998, which has created a substantial antigenic gap between classical swine H1 and human seasonal H1 viruses, to the point that swine became a reservoir of H1 viruses with potential to cause major respiratory outbreaks or even a pandemic in humans as seen in 2009. It is considered, however, that H1N1 in swine in the 1930s arose as a consequence of transmission from humans, or from a common source to the 1918 virus, and that swine viruses from the 1930s are likely to be closely related to viruses circulating in humans at that time.

All adults will therefore have had exposure at some time of their lives to human H1N1 viruses circulating during the last 90 years. Adults born between 1957 and 1977 will not have been exposed to H1N1 viruses as the primary influenza virus encountered in childhood, which may have a bearing on the nature of T-cell memory response and profile of durable subtypic antibody produced. There are no obvious differences, however, in the profile of antibody responses in the cohort of those aged 30–50 years in 2008 using the functional antibody tests described in this report. Elderly adults, i.e. those born in the first decades of the twentieth century, will have encountered 1918-like influenza viruses or close antigenic variants early in life, possibly as a primary infection, and this is likely to be reflected in the measurable antibody response seen to H1N1 2009, which shows closest genetic relationship to the older H1N1 viruses. There are limited data on older H1N1 viruses, derived from study and reconstruction of 1918 viruses, and use of early swine virus isolates from the 1930s. There are no isolates between 1918 and 1930 to help deduce the nature of antigenic variation during this time.

Structural and sequence analysis of H1N1 viruses shows that the HA of pandemic H1N1 2009 virus shares conserved antigenic epitopes with human and swine H1 viruses from the early twentieth century, and that HA from isolates from the 1918 pandemic (A/South Carolina/1/1918) are remarkably close relatives of the pandemic virus, showing only 20% amino acid difference in the antigenic sites, whereas HAs from viruses isolated later, A/Puerto Rico/8/1934 and A/Brisbane/59/2007, differ by 46% and 50%, respectively. 16,17 Recent mouse model experiments have confirmed that prior infection with 1918 H1N1 influenza affords a degree of protection against 2009 pandemic virus antigenic sites on the HA protein of swine lineage influenza viruses that have been preserved from 1918 to the present, and these epitopes may be susceptible to neutralisation by antibodies induced by variants closely related to 1918 influenza. 18 These in vivo experimental data support genetic information about likely relationships between ancestral H1N1 and H1N1 2009.

The proportion of individuals with antibody detectable by MN but not HI (Figure 12a) increased with age and was around 1 : 4 of all those aged 65 + years. As MN measures a larger range of neutralising antibodies than HI, this is consistent with the broad heterosubtypic immunity expected from a lifetime of exposure to influenza A viruses. Comparing the antibody targets of these two tests, HI analysis detects almost exclusively antibodies to the receptor binding site of the HA on the globular head of HA, which is considered to be the main target of neutralising antibodies. MN detects a broader range of neutralising antibodies, which may be directed against more conserved regions of the HA molecule or other viral targets. As a result, MN is a more sensitive and less strain-specific assay than HI,19 although there is generally a good correlation between HI and MN antibody titres for detection of post-infection or vaccination antibody to homologous strains. However, in unexposed individuals, MN is the more sensitive test for detection of cross-reactive antibodies. Individuals showing antibody by MN, but not HI, have been identified frequently (HPA unpublished data). 19,20 While antibody in the baseline is probably caused by previous exposure to seasonal influenza viruses (cross-reactive antibodies, mainly detected by MN, and perhaps longer lasting), antibody levels detected post summer and autumn waves are almost certainly attributable to infection, characterised by recognition of both, strain specific epitopes (detectable by HI, perhaps shorter lasting) as well as conserved epitopes and thus increases the agreement between both functional tests.

The same principles apply also for the older age groups and could explain why we observed the highest levels of discordant results in these groups. The high frequency of positive MN, but negative HI titres indicates the high level of antibody recognising conserved epitopes in these age groups. The agreement between both tests does not increase as a result of the pandemic, as these individuals had a far lower infection rate, indicated by the UK case estimates, numbers of hospitalised/deceased patients and HI data from this study. However, to be more conservative, and as correlates of protection for MN are unknown, we based most of our analysis on HI titres of ≥ 1 : 32. 5

Using the HI correlate of protection of ≥ 1 : 32 suggests that about 25% of adults aged 65 + years should be protected, whereas it may be as high as 67% if any detectable MN antibody is protective. Estimates of the baseline age-specific immunity to H1N1 2009 from transmission models may help determine the likely correlate of protection with each assay, which would be helpful in refining a crude estimate of protection.

Data from our baseline serosurvey are therefore consistent with historical observations and contemporary laboratory experimental work. Cross-reactive antibody titres are highest in birth cohorts that were born during the 1918–57 circulation of H1N1, with the highest titres in those born earlier in that period. However, it is also the case that the oldest age groups will be the most highly vaccinated individuals, consistent with UK national vaccination policy recommending vaccination for all persons over 65 years. A majority of those older than 65 years in our study, birth cohorts from 1910 to 1940, will have received at least one influenza seasonal influenza vaccine, and, more likely, multiple influenza vaccines in the last decade. The vaccines in use in UK vaccination campaigns are all trivalent inactivated vaccines, where the major antigen is the viral HA from influenza A and B strains selected to be representative of circulating viruses. This will include representatives of circulating H1N1 viruses that are antigenically distant from H1N1 2009. There are now reports of prior infection with seasonal influenza A in animal models reducing pandemic virus shedding and transmission. 21 Preliminary data also suggest that seasonal influenza vaccination may confer a degree of cross-protection to other H1N1 strains. 22 In the sample set analysed for this report, seroreactivity to pandemic virus after seasonal influenza vaccination is limited and does not appear to provide significant cross-reactive antibody to H1N1 2009, as illustrated in Table 3. We showed increased seroreactivity to the NIBRG122 virus after receipt of 2005 seasonal vaccine (southern hemisphere) in only 5.8% of all subjects. Similarly, only 7.9% of subjects with clinical diagnosis of ILI prior to the 2009 pandemic showed evidence of seroconversion. Conversely, none of the subjects from the SRI panel seroconverted. Overall, the extent of cross-reactive antibody derived from our own analysis is consistent with work described elsewhere. 20,22 However, the extent to which vaccination in an older population may boost pre-existing antibody and contributes to protection remains to be further explored, and has not been the subject of this work.

Although the data from these assays are useful, and help provide a more detailed explanation of the observed patterns of pandemic attack incidence, they are necessarily pragmatic with a reasonably simple read-out, amenable to high-throughput laboratory work. This provides a population-based approach to analysis of immunity from many different individuals. However, detailed epitope mapping of antibody repertoire at an individual level in older populations is required to deduce the exact reactivity of neutralising antibody reactive in either MN or HI assays for an in-depth understanding of the nature of cross-protection provided in older individuals. This will be fundamental to understanding the pattern of morbidity and mortality seen in all pandemics, where there is sparing of the older adult population, and may have generalisable lessons for vaccine design and the induction of broad cross-reactive immunity. There are different scientific and technical approaches to this, which could include molecular cloning of B cells genes to look at immunoglobulin repertoire23 or the use of phage display to dissect antibody repertoire in individual patients. 24 Both approaches are highly labour intensive, and currently unsuitable for the population-based approach taken here, but are likely to yield important insights in coming years to provide the fundamental detail to explain some of the observations that we and others have made.

Seroincidence estimates

The seroprevalence results for the samples collected after the summer wave in September 2009 were the first to reveal the true extent of H1N1 2009 infection, particularly among school-aged children in London. 8 When compared with the estimated number of clinical cases in this group using the HPA statistical method2 the seroincidence estimates were around 10 times higher. A similar 10-fold scaling factor was estimated from the HPA real-time model. This used the baseline immunity data from the serological survey, together with information on the rate of increase of cases early in the second wave of the pandemic to estimate the age-specific proportion of the population infected in the first wave. 4

Whether the 10-fold difference is due to overestimation of the proportion of individuals with ILI who consulted a health-care professional during the first wave (assumed to be between 20% and 50% in the HPA method) and/or a greater than expected proportion of infections that were asymptomatic or atypical in their clinical presentation cannot be determined on the basis of the serological data alone. In a small household contact study in Hong Kong, of 11 patients with serological evidence of H1N1 2009 infection (fourfold rise in MN titres between acute and convalescent sera) three (27%) had an ILI, although six (55%) had two or more respiratory symptoms. 25 These proportions are similar to those reported for infections with seasonal influenza viruses26 and from a field study in a school in England in May and June 2009 (HPA unpublished data). There is little information, however, on propensity to consult among individuals with ILI in the pandemic in the UK, apart from a small web-based survey that was largely restricted to adults. 27 A large telephone survey in New Zealand suggested that only around 1 in 18 individuals with ILI consulted in the first pandemic wave in that country. 28 This estimate is considerably higher than that used to derive the HPA clinical case estimates.

Our seroincidence estimates show that the second wave of infection that started when the schools re-opened in September 2009 was considerably larger than the first wave that occurred in the summer (Table 14). The relative magnitude of the two waves, as estimated from the serology, is thus more in line with that suggested by the mortality data than the HPA case estimates (Figure 1), which did not take account of likely changes in the propensity to consult over time. There is evidence both from likelihood-based estimates of seroincidence in the second wave and other surveillance data sources3 that the propensity to consult is likely to have been lower in the second than the first wave of infection in the UK.

The difference in magnitude between the two waves varied between regions. In the non-London regions, there was little evidence of infection outside the 5- to 14-year age group in the first wave (estimated seroincidence for this group 6.2%, 95% CI 0.3 to 13.4). In contrast, all age groups outside London showed evidence of infection in the second wave, ranging from 53.1% (95% CI 43.9% to 60.3%) in 5- to 14-year-olds to 15.4% (95% CI 8.7% to 21.4%) in those aged 65 + years (Table 15). The pattern was different in the London region where the first wave was larger than the second (Table 16). The pandemic started earlier in the London region and reached the highest estimated rate of clinical cases/100,000 population of any region in the first wave8 before the national closure of schools took effect and temporarily reduced R to < 1 in all regions. The seroincidence data for the first wave are thus consistent with the regional clinical case estimates using the HPA method. Non-London regions then experienced their major wave of infection when the schools reopened after the summer holidays. The post-second-wave serology results provide some evidence that the final cumulative incidence was higher in London than elsewhere. Given the particular demographic structure of London with a smaller proportion of individuals from the older age groups29 (i.e. those groups with pre-existing immunity), and more contacts due to population density, it can be hypothesised that the higher attack rates in London were due to a bigger R during the two waves. This, together with the likely higher number of imported infections in London than elsewhere, may explain why this region experienced an earlier start of the pandemic and larger overall incidence than other regions.

In addition to regional differences in the timing of the start of the pandemic, there were also differences in timing between age groups, with adults over 24 years in all regions being infected later than those in younger age groups. This is consistent with the key role of children in transmission in the early stages of the pandemic and the social mixing patterns between and within age groups. 30 Clearly, school children are a key target for intervention during a pandemic, especially if the aim is to delay its progression in order to buy time until pandemic strain vaccines become available. The widespread use of antiviral prophylaxis in schools as part of the UK containment policy31 did not appear to be effective in delaying progression of the first wave of infection. While antiviral prophylaxis may have been effective at an individual level1 initial cases in schools were often not identified early enough for prophylaxis to have had a major impact on disease transmission. 32 In contrast, school closures seem extremely effective in reducing transmission as seen by the termination of the first wave when all schools closed for the summer holidays. However, use of national school closure at other unplanned times as a pandemic control measure could result in a considerable economic and social burden. Localised school closures could potentially alleviate the burden on hospital intensive care units that are reaching capacity but a recent modelling study33 shows that, for a range of epidemiologically plausible assumptions, considerable local coordination of school closures would be needed to achieve a substantial reduction in the number of hospitals where capacity is exceeded at the peak of the epidemic. The heterogeneity in demand for intensive care beds means that even widespread school closures are unlikely to have an impact on whether demand will exceed capacity for many hospitals. 33 If school closure is to be used as an intervention strategy in a future pandemic, its deployment may need to be reserved for a more severe strain than the H1N1 2009 virus.

It is not straightforward to disentangle to what the extent the low attack rate in older age groups (and conversely the very high attack rate in younger age groups) is explained by differing social mixing patterns or protective immunity from past infections. There are uncertainties around measuring mixing patterns, partly as a result of their survey origin and partly because it is not clear which type of contacts best describe influenza transmission (i.e. physical or conversational30). There are also uncertainties regarding the interpretation of the high level of baseline cross-reactive antibodies among older persons found in this study in terms of protection against infection and disease. A modelling approach has been used to infer how well mixing patterns derived from the POLYMOD (Improving Public Health Policy in Europe through Modelling and Economic Evaluation of Interventions for the Control of Infectious Diseases) study30 and the observed age-specific immunity profile as described in this study predict the observed pattern of infection. 4 This showed that both factors contribute to reinforce the high attack rate in younger age groups. Some early predictions4 based on scenarios fitted to the HPA clinical case estimates up to October 2009 with a 10-fold scaling factor to derive the number of infections, indicated cumulative attack rates of around 42% in 5- to 14-year-olds and 4% in those aged 65 + years. Even if these numbers need to be revised in the light of the serological data generated in the second wave, these model predictions show that the difference in attack rates between school age children and the elderly can be explained easily by a combination of social mixing patterns and pre-existing immunity.

Study limitations

The serum samples in our study were not obtained as a result of a population screening approach whereby individuals or families are selected at random and asked to provide blood samples for a specific study. While such a method does allow additional clinical and epidemiological information to be obtained with the serum sample, it generally suffers from a low participation rate, which may itself introduce bias and offset the advantages of the random sampling approach. In addition, young children are usually excluded for ethical reasons. For conducting a rapid H1N1 2009 seroincidence survey, the time required to obtain ethics approval and individual patient consent, together with the logistics and costs of such an approach, rendered a population-screening method impractical. The use of serum samples taken from individuals accessing health care for clinical reasons unrelated to a recent illness suggestive of influenza provided a convenient alternative method.

Construction of an HPA annual serum archive from residual aliquots of samples submitted to microbiology laboratories for screening or diagnostic testing has been in operation for over two decades, having been originally established to monitor the impact of the combined measles, mumps and rubella vaccine on age-specific population immunity. 34 The HPA archive has been extensively used for other seroepidemiological studies35 and has proven particularly useful for infections with a high incidence for which exposure is largely age dependent rather than determined by specific behavioural factors (such as for HIV) or associated with particular ethnic groups (such as hepatitis B). Because of the need for rapid generation of incidence data for H1N1 2009, serum collection was extended to chemical pathology laboratories – a new source for seroepidemiological studies. While there were no significant differences in H1N1 seroprevalence between samples from microbiology and chemical pathology laboratories in the logistic regression model (Table 13), patients undergoing regular chemical pathology testing may be more likely to have underlying morbidities (such as chronic cardiac, renal or respiratory conditions) than the general population. This could bias the results if such conditions affect the likelihood of being infected with the H1N1 2009 virus or the likelihood of having been vaccinated with the H1N1 2009 pandemic vaccine. Unless the underlying clinical conditions have a major impact on mixing patterns then bias in the estimates of H1N1 2009 infection through differential exposure would be unlikely. However, these clinical conditions are indications for seasonal and pandemic influenza vaccination, which is a potential source of bias that needs to be considered. As discussed earlier, it is unlikely that prior seasonal influenza vaccination will have generated cross-reactive antibody but the roll-out of pandemic stain vaccine in the UK for high-risk individuals from early November 200936 and for all children under 5 years of age from January 201037 may have biased the analysis, at least in the samples taken from November onwards.

The pandemic vaccine uptake rates by age group in England, derived from database extracts from the 96 general practices enrolled in the RCGP network,38 are shown in Figure 14. Given the relatively high uptake rates in the 65 + age group (26% overall and 41% in high-risk groups), and to a lesser extent in the 45- to 64-year age group (11% overall and 41% in risk groups) some of the increase in the prevalence of H1N1 2009 antibodies in these age groups between the 2008 baseline and the post-second-wave sera is likely to have been due to vaccination, even without selective inclusion of sera from high-risk individuals targeted for vaccination. Vaccine uptake among those aged between 5 and 45 years was much lower overall (< 5%) due to the lower proportion with underlying clinical conditions in these age groups, although coverage in risk groups was relatively high (30%, 19% and 27% for the groups aged 5–14, 15–24 and 25–44 years, respectively). The extent to which vaccination may have contributed to the change in antibody prevalence compared with the baseline in the sera collected from November onwards is therefore difficult to assess. However, even if sera in the 5- to 14-year-olds were exclusively from those in risk groups, the vaccine gave 100% seroconversion, and vaccination only occurred in persons not already infected, the increase in seroprevalence between the first and second wave (51.4%, as shown in Table 14) in this age group is substantially higher than the 30% observed uptake in 5- to 14-year-olds in a risk group. For those aged < 5 years, vaccine was being delivered from January 2010 onwards (while the post-second-wave sera were being taken) and was targeted at all children not just those in risk groups so increases in seroprevalence would be expected from vaccination. It is perhaps surprising therefore that there was evidence of a decline in the proportion with HI titres ≥ 1 : 32 after January in this age group (Figure 7).

FIGURE 14.

Percentage of populations vaccinated with pandemic strain vaccine by age group and week, 2009 and 2010 (data from RCGP).

In estimating cumulative incidence across the two waves of infection, we assumed that HI and MN antibodies developed in response to infection remain at a titre of ≥ 1 : 32 or ≥ 1 : 40, respectively, for around a year, such that that individuals infected early in the first wave (e.g. school-aged children in London in May 2009) would still be seropositive if tested in say March/April 2010. The data on antibody titres in confirmed cases did not allow this assumption to be tested, as there were few samples taken later than 90 days after onset and few overall from children. It is possible that the decline in proportions with HI titres ≥ 1 : 32 in the March 2010 sera, which was statistically significant only in London, may have been the result of waning antibody levels following infection in the first or early in the second wave in this region.

Ideally, to generate information on the serological response in laboratory-confirmed cases of H1N1 2009 infection, a cohort would have been prospectively recruited and those with a PCR-confirmed infection would be followed up with sequential serology samples. Such an approach would be expensive and time consuming and there is no reason to assume that our method of using single serum samples taken from different individuals at various time after confirmation of infection would have produced biased results. The serum samples from confirmed cases that were used to estimate the temporal distribution of the seroconversion interval distribution were obtained from a variety of sources. Many were obtained from the follow-up of the first few hundred confirmed cases and their contacts (the so called FF100 database1). These were actively sought to try and identify seroconversions in contacts of confirmed cases to derive secondary attack rates. Although few paired samples were obtained it is difficult to envisage why those that were obtained from PCR-positive contacts or index cases were somehow biased with respect to the serological response they developed. Other samples were obtained from subjects investigated as part of outbreaks while some were individuals for whom the clinician decided that a serological test for H1N1 2009 was indicated. The distribution for the seroconversion intervals obtained with this approach was consistent with what would be expected for seasonal influenza.

Other limitations of our method of estimating incidence by measuring differences in prevalence between two time points are that it does not take account of the variable time to seroconversion between individuals and the fact that a small proportion of infected individuals do not appear to seroconvert by methods used in this study. It also relies on grouping by calendar month yet incidence may be changing rapidly over this time period. The likelihood-based estimation was developed to overcome some of these limitations and allowed the generation of a continuous cumulative incidence curve by region and age group over the first and second waves. However, to accommodate complexities such as vaccination, or a non-random serum sampling strategy with the potential for oversampling in risk groups for whom vaccination was recommended, or changes in the propensity to consult over time, a more complex set of parameters would need to be estimated. Additionally, the method can incorporate other data sets, such as the vaccine uptake data by age and risk groups or results from vaccine immunogenicity trials. For this, Markov Chain Monte Carlo (MCMC) could be used with the combined likelihood of the different data sets to draw jointly a sample from the parameter space and then the parameters can be evaluated from the sampled distribution. Such a model could also incorporate the effect of waning antibody levels if this is shown to be important using additional data on antibody persistence after infection or vaccination.

Our study did not provide the timely incidence estimates to inform the parameterisation of ‘real-time’ predictive models as originally envisaged, due to the necessary lag time in H1N1 2009 assay development and, to a lesser extent, the collection and testing of the seroincidence samples. However, had the pandemic been with an H5N1 virus for which serological assays had already been developed then generation of baseline data on the age-specific prevalence of cross-reactive antibodies and insight into the incidence of infection would have been obtained more rapidly. Indeed the HPA serum archive has already been used to measure the prevalence of cross-reactive antibodies to the H5N1 virus. Nevertheless the data generated on the prevalence of cross-reactive antibodies to H1N1 2009, which was available prior to the second wave, did assist in the parameterisation of the HPA real-time model used to evaluate the likely impact of vaccination. 4 It also helped validate the scaling factor used in that model to convert the HPA clinical case estimates to infections. 8 For future pandemics, the availability of new serological tools that can detect incident infection in a single sample (e.g. analogous to an Immunoglobulin M test) rather than relying on changes in seroprevalence over time, and the availability of non-invasive techniques, such as oral fluid testing, would greatly facilitate the rapid generation of seroincidence data through a random population-screening approach.

Finally, our study was geographically limited to England. As the pandemic was UK wide, although with some differences in timing and magnitude,39 it would have been informative if a pan-UK study had been undertaken that included Scotland, Wales and Northern Ireland. A proposal for a similar seroepidemiological study was funded in Scotland, for which we shared our protocol and testing standard operating procedures (SOPs), and work is under way to provide data to compare with our study, although is not available at present. Seroepidemiology from Northern Ireland would also have been useful to compare with England, particularly with the different timing and impact of first wave in Northern Ireland. Given the difference between the UK and other European countries in the timing of the first wave of infection, useful epidemiological insights might have been gained if comparable studies had been conducted in those countries. Although serological data are emerging from a range of European Union (EU) countries, the complexity of health systems and variability of studies undertaken makes developing a regional European picture of impact of the pandemic unfeasible at present.

MN

The microneutralisation (MN) will be performed in 96-well format according to previously described protocols [Nicholson KG, Colegate AE, Podda A, Stephenson I, Wood J, Ypma E, et al. Safety and antigenicity of non-adjuvanted and MF59-adjuvanted influenza A/Duck/Singapore/97 (H5N3) vaccine: a randomised trial of two potential vaccines against H5N1 influenza. Lancet 2001;357:1937–43] and standard operating procedures (SOPs) developed at the Respiratory Virus Unit (RVU).

HI

The principle of the haemagglutination inhibition (HI) test is based on the ability of specific anti-influenza antibodies to inhibit haemagglutination of red blood cells (RBCs) by influenza virus HA. The sera to be tested have to be previously treated to eliminate the non-specific inhibitors and the anti-species HAs. The experiment will be performed in accordance to protocols and SOPs established by the RVU.

August 2009

September 2009

October 2009

November 2009

December 2009

January 2010

February 2010

March 2010

April 2010

Background

These notes set out the main considerations when seeking to undertake seroepidemiological studies to provide real-time rapid response to an emerging infection, such as pandemic influenza. Under these circumstances, serological data provide important information, such as levels of cross-protective immunity in different population groups and incidence rates of infection.

A seroepidemiological study can provide information on:

1. age-specific incidence as the disease emerges in the population

2. prevalence of existing cross-protective immunity

3. cumulative prevalence to inform future incidence predictions, for example through disease modelling where seroepidemiological data can be combined with age-specific morbidity data to predict the likely burden of illness and the impact of a novel infection such as pandemic influenza.

These data can be used to complement other descriptive epidemiological data or to inform disease transmission models used to predict the future course of the pandemic.

The requirements are different from those when setting up a seroepidemiological programme, the main focus of which is to monitor levels of immunity to particular diseases within a population to assist in evaluation of the impact of a vaccination programme or to inform the need for other public health interventions. 1,2

For any seroepidemiological study, it is essential that recent, age-stratified baseline sera are available to inform the interpretation of data. Such sera cannot be collected retrospectively so long-term investment by member states is essential to ensure access to population-based stored sera. The absence of recent baseline sera is a significant limiting factor in the interpretation of data for newly emerging infections. For example, knowledge of the baseline prevalence of antibodies to pandemic influenza A/H1N1 in 2009 was essential for understanding the epidemiology and informing vaccine policy, as it indicated that lower clinical attack rates in the elderly were the result of pre-existing cross-reactive antibody, thereby reducing the clinical benefit and cost-effectiveness of delivering vaccine to this group. This required access to sera obtained prior to the pandemic.

It is essential that appropriate personnel (microbiologists, epidemiologists, mathematical modellers and statisticians) are involved in the establishment of seroepidemiological studies from the outset.

When considering region-wide seroepidemiology (e.g. EU wide) it is necessary to form teams with broad representation, and establish a limited network of laboratories or a central laboratory for testing to ensure maximum generalisability. Comparisons between countries may be confounded by differences in assay performance, sensitivity and specificity. 3,4

References

Annex 1: Seroepidemiology Unit user guide

(PDF download)

Annex 2: laboratory methods