The birth of modern epidemiology is often attributed to John Snow's famous investigation of the 1854 cholera epidemic in London, and his identification of the Broad Street pump as the most important node in the cholera transmission network [
1,
2]. More than 150 years later, we still do not know what factors triggered the enormous cholera outbreaks in London in the summers of 1832, 1849, 1854 and 1866. In addition to the intrinsic interest of identifying the mechanisms of historical disease invasions, improving our understanding of cholera specifically is important because it remains a serious public health concern in areas where clean water is unavailable [
3]. The recent cholera epidemics in Angola in 2006 [
4], Zimbabwe in 2008–2009 [
5] and Haiti in 2010 [
6] are stark examples.
Previous studies of nineteenth century cholera have focused on the ‘great’ cholera years while paying little attention to the years between severe outbreaks [
7–
9]. Here, we consider the great cholera years in the context of London's weekly mortality over the course of the nineteenth century as a whole. Cholera deaths were recorded in the Weekly Returns of the Registrar General's Office from 8 January 1842 to 28 December 1901 (the date on which the last cholera death was reported in London). We digitized these and earlier cholera records from the London Bills of Mortality to obtain a contiguous weekly record of cholera deaths for 77 years from 24 August 1824 to 28 December 1901.
displays the London cholera data in two ways. a shows the weekly time series, while b shows an intensity plot of week-of-year against year to bring out the pattern of seasonality over the years. The three striking features highlighted above are readily apparent: (i) London's cholera epidemics were strongly seasonal and most intense in the summer, (ii) typical outbreaks were far milder than the devastating outbreaks in the summers of 1832, 1849, 1854 and 1866, and (iii) atypically timed outbreaks occurred in the few months preceding the major outbreaks in 1832, 1849 and 1854. The data clearly separate into regular, mild summer outbreaks, together with outliers corresponding to the non-summer outbreaks and the great cholera years. These features persist when cholera deaths are normalized to account for changes in population size and reporting coverage over the century (see and the electronic supplementary material).
What factors might have been responsible for the unusual timing of the non-summer outbreaks in 1832, 1848 and 1853, and the unusual severity of the outbreaks in the following summers? The most obvious hypothesis is that these features of the London cholera time series resulted from the introduction of new cholera strains into the city at random times of year. It is natural to expect that a non-summer introduction would result in an initial outbreak, with severity tempered by the season [
10], followed by a severe outbreak in the summer when environmental conditions most strongly promote cholera transmission. The non-summer outbreaks thus appear to have ‘heralded’ the arrival of new cholera strains in London in 1832, 1848 and 1853. The absense of a ‘herald wave’ before the major epidemic in 1866 may simply reflect the invasion of a new strain coincidentally near the start of the normal cholera season that year.
The idea of herald waves has largely been confined to influenza [
11], where it has been discussed extensively in relation to the 1918 [
12–
14] and 2009 [
15,
16] pandemics. To our knowledge, herald waves have never been described for cholera. Our finding of herald waves for London cholera suggests that herald waves may be a common feature of seasonal diseases.
We note that the identity and origins of new cholera strains in the nineteenth century are not known. Current evolution of new cholera strains appears to be facilitated by horizontal gene transfer among strains in different serogroups ([
17]; see the electronic supplementary material). This mechanism might also account for the invasion of antigenically novel strains in the nineteenth century. Cyclical replacement of the predominant cholera serotype has also been observed [
18], and may be relevant for London cholera.
We used a simple mathematical model to verify theoretically that a non-summer introduction of a new cholera strain can result in a herald wave followed by a severe outbreak in the summer. The model extends the classical ‘susceptible–infectious–recovered’ (SIR) framework to include a water compartment (W), with transmission occurring through both person–person and person–water–person pathways ([
19]; see appendix A for model equations).
As an example, compares the London cholera mortality time series from 1845 to 1852 (
a) with an SIWR simulation (
b). Assuming the transmission rate in the water varies seasonally (peaking in the summer, as in
c), simulated cholera time series resemble the typical pattern in London, with mild annual summer outbreaks (as in the period 1845–1848 in
b). The arrow in
b indicates the time at which we introduced a new cholera strain into the SIWR model (the end of September 1848). The population is completely susceptible to the new strain, resulting in an initial outbreak when the strain is introduced. However, owing to the season [
10], disease transmissibility is waning at the time of introduction. Eventually transmission decreases past the point where an outbreak can be sustained (and the initial herald wave terminates), but the new cholera strain persists at low levels in the water (
d). When transmissibility from the water rises again the following summer, it triggers an unusually severe epidemic owing to the large number of susceptible individuals (simulation details given in appendix A).
While the strain invasion hypothesis is simple—and appealing from the point of view of parsimony—many other explanations are possible, including misdiagnosis involving other diarrhoeal diseases. A direct test of the strain invasion hypothesis would require genetic information from the circulating cholera strains. Relevant tissue samples from patients who died of cholera do exist in museum collections, and recent advances in the recovery and sequencing of DNA as well as the reconstruction of complete genomes from fossil materials [
20,
21] make sequencing substantial portions of the genomes of nineteenth century cholera strains a realistic goal (see the electronic supplementary material).
Given the fame and historical importance of the four major London cholera epidemics in the nineteenth century, it is surprising that the herald waves we have identified here have not been highlighted previously. Unravelling the mechanisms behind these herald waves will deepen our understanding of the evolutionary and ecological history of this important disease, and in turn help us understand the factors underlying severe cholera outbreaks in modern times. Our study of London cholera also suggests that herald waves may occur for more diseases than has been previously realized, and emphasizes the need for further work examining the relationship between the timing and magnitude of seasonal outbreaks [
10]. The systematic digitization of lengthy historical records of disease incidence and mortality will be invaluable for this endeavour. In particular, evidence that herald waves have preceded major epidemics of other diseases may be hidden in untapped historical sources.
This article has been 
B