Background
Lessons from past influenza pandemics, including the great pandemic of 1918-19 [
1,
2] can enhance understanding of later pandemics [
3‐
5], such as the recent pandemic of H1N1 2009 [
4,
6‐
11]. Indeed, part of the genetic sequence of the H1N1 virus [
12] from the 1918-19 influenza pandemic lives on in H1N1 2009 [
9]. Further insights from immunology [
8,
13,
14], from animal studies using reconstituted viruses [
15], and from epidemiological analyses and modelling of past and current outbreaks and future scenarios [
3,
16‐
27] will add to understanding.
The 1918-9 pandemic was characterised by high mortality, particularly in isolated or disadvantaged populations [
1,
13,
28]. In urban populations, mortality rates were relatively greater amongst young adults [
1,
2,
16,
17], probably because older adults were protected by persistent immunity from a related virus that had disappeared by 1890 [
4,
5,
17,
18], while children were protected by innate immune mechanisms such as those mediated by interferon [
17]. The out-of-season onset in 1918 [
1‐
3,
19] likely reflected greater population susceptibility to that novel H1N1 virus, while the multiple waves of infection could have been due to waning of short-term immunity [
3,
5,
21], antigenic drift of the virus [
3,
5,
19,
21], social distancing [
22], &/or seasonal effects [
5,
19,
29].
The pandemic of 1918-19 killed at least 0.2% of persons in most affected populations, and as many as 20% in some areas [
1,
2,
16]. Some of this variation between populations in 1918-19 mortality has been explained by poverty [
16], possibly mediated via overcrowding and malnutrition [
5,
16]. The very high attack-rates and mortality rates in places such as Alaska and Western Samoa [
1,
2,
17] and amongst Aborigines in remote Australia [
30] in 1918-19, and on the isolated island of Tristan da Cunha in 1971 [
3,
31], have led to suggestions that such isolated populations were vulnerable because they had escaped regular infection with seasonal influenza viruses and were thus left with little or no immune protection against the pandemic virus [
3,
5,
13,
17,
31]. In more urbanised communities, pandemic behaviour was unusual in other ways. For example, separate waves of influenza were clearly demarcated in summer, autumn and winter in 1918-19 in England and Wales [
1,
2,
13]. Despite those three waves of potential exposure, many persons did not report symptoms of influenza in any wave [
1,
2,
13] (see Table
1). It has not been clear whether such persons were unexposed, whether infections were asymptomatic or unreported, or whether persons were protected by innate immunity, or by residual cross-immunity from other influenza viruses circulating before 1918 [
2,
3,
5,
17,
18].
Table 1
Populations and observed proportions affected in summer, autumn & winter waves in 1918-19 pandemic
South Shields | 462 | 844.2 | 26.0 | 51.9 | 67.1 | 2.2 | 2.2 | 6.5 | 0 |
Leicester | 4619 | 719.9 | 62.1 | 131.8 | 69.9 | 3.0 | 4.8 | 8.0 | 0.4 |
Wigan | 1075 | 774.0 | 40.9 | 73.5 | 108.8 | 0 | 0 | 1.9 | 0.9 |
Newcastle | 4461 | 814.4 | 52.5 | 46.6 | 73.1 | 0.4 | 8.7 | 3.8 | 0.4 |
Manchester | 4686 | 747.1 | 130.8 | 83.7 | 15.6 | 14.3 | 5.5 | 2.3 | 0.6 |
Blackburn | 1284 | 785.0 | 75.5 | 56.1 | 64.6 | 5.5 | 4.7 | 7.8 | 0.8 |
Widnes | 3417 | 696.5 | 113.5 | 77.8 | 99.5 | 4.1 | 6.1 | 2.3 | 0 |
London police | 746 | 749.3 | 61.7 | 144.8 | 32.2 | 5.4 | 0 | 6.7 | 0 |
Cambridge Uni | 1766 | 457.0 | 206.7 | 208.4 | 73.6 | 18.7 | 9.6 | 21.5 | 4.5 |
Clifton College | 451 | 232.8 | 153.0 | 84.3 | 188.5 | 20.0 | 157.4 | 135.3 | 28.8 |
Haileybury | 515 | 302.9 | 227.2 | 93.2 | 205.8 | 60.2 | 42.7 | 48.5 | 19.4 |
Finchley School | 1224 | 550.7 | 90.7 | 312.9 | 23.7 | 14.7 | 4.1 | 3.3 | 0 |
ALL
| 24706 | 703.1 | 96.5 | 105.0 | 67.5 | 8.1 | 9.3 | 8.9 | 1.6 |
Until recent work [
3,
32], based on data from 1918-19, there was even uncertainty about whether an attack of influenza in an early wave of the pandemic protected an individual in a later wave, as would be expected if the viruses in each wave were similar. Indeed, the English data in Table
1[
2] were puzzling even to FM Burnet, the leading Australian virologist, when he and Clark reviewed the 1918-19 pandemic evidence in 1942 [
13].
It occurred to us that if there were immunity in some persons before the summer wave [
3,
5,
13,
17,
24], &/or if some infections were asymptomatic [
3,
20], this could explain the inconsistent evidence for the attack-rate in a later wave being reduced in individuals reporting symptoms in an earlier wave (Table
2). Accordingly, we now report our innovative modelling to test that hypothesis.
Table 2
Observed odds ratios (OR) and 95% confidence intervals to test for evidence of immune protection from wave to wave.
South Shields | 1.199 (0.045 - 11.364) | 0.937 (0.036 - 8.735) | 1.508 (0.292 - 6.549) |
Leicester |
0.292
(0.158 - 0.533) | 0.871 (0.521 - 1.446) |
0.655
(0.437 - 0.979) |
Wigan | 0.266 (0.011 - 2.186) | 0.174 (0.007 - 1.419) | 0.284 (0.061 - 1.090) |
Newcastle |
0.258
(0.071 - 0.816) | 1.945 (1.284 - 2.936) | 0.959 (0.536 - 1.692) |
Manchester | 0.972 (0.709 - 1.329) | 1.976 (1.182 - 3.286) | 1.267 (0.635 - 2.481) |
Blackburn | 1.033 (0.406 - 2.522) | 0.782 (0.290 - 1.996) | 1.729 (0.763 - 3.821) |
Widnes |
0.340
(0.175 - 0.648) |
0.397
(0.230 - 0.677) |
0.219
(0.091 - 0.505) |
London police | 0.449 (0.117 - 1.515) | 0.000 (0.029 - 2.933) | 1.125 (0.324 - 3.610) |
Cambridge Uni |
0.248
(0.165 - 0.371) |
0.439
(0.261 - 0.734) | 0.915 (0.604 - 1.380) |
Clifton College | 0.302 (0.164 - 0.551) | 1.055 (0.657 - 1.687) | 1.756 (1.044 - 2.946) |
Haileybury | 1.059 (0.634 - 1.766) |
0.337
(0.199 - 0.567) | 0.945 (0.550 - 1.620) |
Finchley School |
0.282
(0.152 - 0.518) | 1.241 (0.368 - 3.841) |
0.230
(0.060 - 0.774) |
ALL
|
0.621
(0.529-0.729) | 1.105 (0.946-1.291) | 0.972 (0.831-1.138) |
Discussion
Our modelling results using 1918-9 data support earlier suggestions [
3,
17] that the spread of pandemic influenza can be limited by pre-existing immunity, probably resulting from prior exposure to seasonal influenza [
34]. Furthermore, the waning of prior immunity likely contributed to the recurrent waves of influenza that characterised the 1918-9 pandemic in urban populations. These modelling inferences are necessarily tentative, as they cannot be supported by studies of immune mechanisms in those historical populations. Nevertheless, our modelling approach is innovative, biologically plausible and uses modern estimation procedures. (Additional file
1 provides further details of the methods and potential limitations of our approach.)
Our paper is able to make strong inferences about asymptomatic infections, immunity,
R and
R
0
without having to estimate or guess, as is usually the case, the serial interval of influenza infection [
3]. This was possible because we had data on the final size of each of the three waves in 12 sub-populations, and because we assumed homogeneous mixing within each sub-population, to underpin the (deterministic) "final-size" equation used to link the attack rate to the parameters (see methods and Additional file
1). Although the assumption of homogeneous mixing can only be an approximation, it seems reasonable, and is frequently made. Furthermore, in the supplementary information (Additional file
1) we show that our model conclusions are robust to effects arising from the simplest form of social distancing, although we cannot exclude more complex forms of social distancing as an ancillary explanation for the wave-like behaviour of influenza [
22]. However, social distancing alone cannot explain why some persons had repeat attacks from wave to wave. In ongoing work, we are relaxing the assumption of homogeneous mixing, and testing the robustness of our conclusions against more complex models of social distancing.
Our results show that
R
0
estimates varied somewhat between populations, and tended to be greater in schools, as would be expected with higher mixing rates. In the results presented, we did not allow for systematic variation of
R
0
from wave to wave, as in other analyses (not shown) we found that this did not lead to a significant improvement in model fit. Our assumption of an
R
0
that did not vary between waves also means that we have disregarded the potential effects of seasonality on
R
0
and transmission [
29]. However, our work suggests that the attack-rate is related more directly to the proportion susceptible
(Z), and to population-specific mixing as measured by
R
0
, and that seasonal effects may be of lesser importance. Indeed, we believe that the summer onset of the first wave in England in 1918 was because the antigenic novelty of the new virus, by increasing
Z, had over-ridden the seasonal effect.
There is precedent for our view that cross-reactive immunity induced by prior exposure to a different subtype of influenza can provide partial protection against a new pandemic strain [
3,
5,
18,
19]. Such heterosubtypic immunity is well documented in mouse models [
35,
36], while the evidence from human studies, although inferential, is supportive. Indeed, the very replacement of H1N1 by H2N2 in 1957, and of the latter by H3N2 in 1968 [
19] provide strong circumstantial evidence for the importance of heterosubtypic immunity at the population level. Cross-immunity of short duration between different influenza strains has also been invoked to help explain the apparent constraints on the evolutionary diversification of influenza A [
21]. More directly, Slepushkin reported in 1959 [
37] that persons with symptoms during the H1N1 influenza in the spring of 1957 were less likely to be symptomatic in the summer when the new H2N2 influenza appeared (odds ratio = 0.418, 95% confidence interval 0.304-0.575); in the later autumn wave of H2N2, the level of protection had declined (OR = 0.625, CI = 0.530-0.737). Epstein [
34] re-analysed viral isolation data from the Cleveland family study before and after the arrival of H2N2 in 1957, and found that adults known to be infected by H1N1 over the period 1950-57 were less likely to be infected with H2N2 in 1957 (OR = 0.294, CI 0.01-3.07), although the difference was not significant because of the small numbers. In the Seattle family study over the period 1975-79, coinciding with the return of H1N1 [
38], the age-related decline in attack-rate was only partly explained by hemagglutination inhibition (HI) antibodies. Adults were rarely infected with H1N1 regardless of HI antibody titre, possibly because of cumulated heterosubtypic immunity from recent exposures to H3N2, although older adults in the study could have been protected by H1N1 memory from exposures prior to 1957, before H1N1 was replaced by H2N2. In 1985, Sokoguchi et al [
39] reported strong cross-protection between H3N2 and H1N1 in almost contemporaneous outbreaks in Japanese schools in 1978 (OR = 0.059, CI = 0.019-0.131 for high school students, and OR = 0.154, CI = 0.076-0.309 for younger students). Such strong cross-protection when exposures to different subtypes were separated by only a few days or weeks [
39] is to be contrasted with the weaker protection reported when sequential exposures were more widely separated in time [
37], suggesting that at least some components of cross-protection can fade rapidly, consistent with our interpretation of the 1918-19 data.
What are the mechanisms of heterosubtypic immunity? Studies in mice and other experimental animals have implicated mucosal antibodies, CD4 and CD8 T-cells, and B cells [
35,
36]. Cytotoxic (CD8) T-cells reacting with conserved epitopes on internal viral proteins are of particular importance in eliminating virus-infected cells, thereby reducing the severity and duration of infection [
19,
34‐
36,
40‐
44]. HLA-restricted CD8-mediated cytotoxic activity is also widespread in humans [
36,
43‐
45]. For example, cytotoxic cells from most healthy subjects in UK and Vietnam recognise epitopes of seasonal influenza, as well as similar epitopes of H5N1 avian influenza [
45]. McMichael and others have shown that specific CD8 cells reduce viral shedding and duration of infection in people, and that cytotoxic activity fades over several years without re-exposure [
43,
44]. In young children, cellular immune responses induced by live-attenuated influenza vaccine appear to protect against laboratory-confirmed influenza [
40].
Such collateral evidence supports our view [
3,
5] that in 1918-19 many people in cities were at least temporarily protected from pandemic influenza by pre-existing heterosubtypic immunity, presumably induced by recent exposure to seasonal influenza. We propose that pre-existing heterosubtypic immunity was often short-lived, and that immunity to a new strain or subtype also required several exposures before becoming more permanent. For example, it is possible that heterosubtypic protection antedating wave 1 was mediated by CD8 T-cells, which could fade over time in persons not exposed, or when exposure did not result in a significant viral load. The primary antibody response in persons exposed to larger viral loads in wave 1 could have faded in some persons before wave 2. By wave 3, immunity could have been consolidated in those with several exposures through the production of longer-lived antibody of IgG class. This more permanent protection would have helped to defer the next outbreak to the influenza season of 1920, and started the transition from pandemic to seasonal behaviour [
19,
46].
How do our findings relate to the 2009-10 pandemic caused by the H1N1 2009 virus of swine origin? Despite changes in social conditions since 1918, published estimates of the effective reproduction number (
R) for the new swine flu are in the range 1.2-3.1 [
47‐
49], with the larger estimate relating to transmission between minors in Japan [
48]; these results are consistent with our findings from 1918-19, including higher rates of transmission for "schools" (Table
4). In our results we draw an important distinction between the higher estimates for
R
0
and the lower estimates for
R at the start of the outbreak. This difference reflects the effect of prior immunity in moderating the spread of pandemic influenza in 1918-19 [
2,
3,
5,
24]. Unfortunately, although most influenza modellers have estimated
R, some have reported it or used it as
R
0
; we suggest [
11] that this could systematically under-estimate [
25‐
27] what the rate of spread of influenza would be in more fully susceptible populations, as in isolated locations such as Tristan da Cunha in 1971 [
3,
31] or in sequestered schools such as Saffron Walden in 1918-9 [
2,
3].
Although H1N1 2009 swine flu shows the pandemic signature of a relatively greater mortality in young adults [
6,
50], aggregate influenza mortality in 2009 [
6,
10] seems lower than that from seasonal influenza, which typically affects the elderly[
19]. Such observations support the growing consensus that the H1N1 2009 virus has also been spreading in partially immune populations [
10,
11,
47,
51‐
53]. However, pre-existing cross-reactive antibodies to H1N1 2009 seem confined to older persons, presumably directed against epitopes not present in the recent H1N1 seasonal virus [
8,
51]. CD8+ T-cells directed against conserved influenza epitopes, which would resolve infections early, could help to explain the constrained spread of H1N1 2009 even in persons without neutralising antibody [
54]. If cross-immunity is limiting the rate of spread of the H1N1 2009 virus in the same way as for the 1918-19 virus [
3,
5,
11], and if that cross-protective immunity is also short-lived, there is a risk of repeat pandemic waves in 2010. Furthermore, it is possible that without vaccination, populations escaping early infection with the pandemic virus will experience more rapid spread or greater disease severity when eventually infected. Fortunately, trial results suggest that a single dose of pandemic vaccine can induce ostensibly protective levels of antibody, possibly by building on cross-reactive immune memory from prior exposures to seasonal H1N1 virus or vaccine [
52,
53].
Competing interests
JDM was an expert witness on the epidemiology of influenza in a matter before the Supreme Court of Victoria in 2007-8. The parties to that case have had no influence on, no involvement in, and have made no financial contribution to this work. JDM, JMcC, JMcV, and EMcB have provided advice on influenza matters at various times to the Australian Government, or to the Victorian Government.
Authors' contributions
JDM conceived the project, wrote code, identified and analysed data, and drafted the manuscript; EMcB oversighted the Bayesian framework and hyperparameters; JMcC improved the code; all authors reviewed the ideas, methods, preliminary results, discussion and draft manuscript; all authors read and approved the final manuscript.