Top Banner
Modelling of potential pandemics Christophe Fraser MRC Centre for Outbreak Analysis and Modelling WHO Collaborating Centre for Infectious Disease Modelling NAS Symposium on Gain of Function Research Session 4: Potential Benefits of GOF Research I: Surveillance, Detection and Prediction
13

Session 4: Modeling of potential pandemics

Jul 18, 2016

Download

Documents

earthandlife

Christophe Fraser's presentation on modeling potential pandemics.
Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Page 1: Session 4: Modeling of potential pandemics

Modelling of potential pandemics

Christophe Fraser

MRC Centre for Outbreak Analysis and Modelling WHO Collaborating Centre for Infectious Disease Modelling

NAS Symposium on Gain of Function Research Session 4: Potential Benefits of GOF Research I: Surveillance, Detection and Prediction

Page 2: Session 4: Modeling of potential pandemics

Experience from past and current epidemics

the cell cortex is a rate-limiting step for forcegenerator function. Given that cortically lo-calized G!i and Pins also dictate spindlepositioning in Drosophila (10), we proposethat modulation of cortical G! signaling togenerate defined pulling forces on astral mi-crotubules is a conserved mechanism totranslate polarity cues into appropriate spin-dle positioning.

References and Notes1. J. A. Knoblich, Nature Rev. Mol. Cell Biol. 2, 11(2001).

2. P. Gonczy, Trends Cell Biol. 12, 332 (2002).3. J. Pellettieri, G. Seydoux, Science 298, 1946 (2002).4. K. J. Kemphues, J. R. Priess, D. G. Morton, N. Cheng,Cell 52, 311 (1988).

5. S. W. Grill, P. Gonczy, E. H. Stelzer, A. A. Hyman,Nature 409, 630 (2001).

6. M. Gotta, J. Ahringer, Nature Cell Biol. 3, 297 (2001).7. F. Yu, X. Morin, Y. Cai, X. Yang, W. Chia, Cell 100, 399(2000).

8. M. Schaefer, A. Shevchenko, J. A. Knoblich, Curr. Biol.10, 353 (2000).

9. M. Schaefer, M. Petronczki, D. Dorner, M. Forte, J. A.Knoblich, Cell 107, 183 (2001).

10. Y. Cai, F. Yu, S. Lin, W. Chia, X. Yang, Cell 112, 51(2003).

11. P. Gonczy et al., Nature 408, 331 (2000).12. Examination of the genome sequence and sequencingof full-length cDNAs revealed that gpr-1 (WormBasecode F22B7.13) and gpr-2 (C38C10.4) are both ex-pressed and that they result from a recent geneduplication event. We found a single gpr homolog inthe available C. briggsae genome (CBG10100), whoseinactivation caused a phenotype indistinguishablefrom that observed in C. elegans.

13. B. Zhang et al., Nature 390, 477 (1997).14. K. Colombo, P. Gonczy, unpublished data.15. B. Etemad-Moghadam, S. Guo, K. J. Kemphues, Cell83, 743 (1995).

16. L. Boyd, S. Guo, D. Levitan, D. T. Stinchcomb, K. J.Kemphues, Development 122, 3075 (1996).

17. S. Guo, K. J. Kemphues, Cell 81, 611 (1995).18. T. J. Hung, K. J. Kemphues, Development 126, 127(1999).

19. S. Strome, W. B. Wood, Cell 35, 15 (1983).20. K. J. Reese, M. A. Dunn, J. A. Waddle, G. Seydoux, Mol.Cell 6, 445 (2000).

21. L. De Vries et al., Proc. Natl. Acad. Sci. U.S.A. 97,14364 (2000).

22. R. J. Kimple et al., J. Biol. Chem. 276, 29275 (2001).23. M. Natochin, K. G. Gasimov, N. O. Artemyev, Bio-chemistry 40, 5322 (2001).

24. See supporting online material.25. Average peak velocities in "m/s (#SEM at the 0.95confidence interval) were as follows (A, anterior spin-dle pole; P, posterior spindle pole): gpr-1/2(RNAi), A:0.20 # 0.03, P: 0.18 # 0.04, n $ 15; goa-1/gpa-

16(RNAi), A: 0.21 # 0.04, P: 0.22 # 0.03, n $ 15;par-3(it71) gpr-1/2(RNAi), A: 0.18 # 0.04, P: 0.18 #0.03, n $ 13. Values for par-3(it71) were A: 0.89 #0.11, P: 0.90 # 0.07, n $ 20 (5).

26. K. G. Miller, J. B. Rand, Genetics 156, 1649 (2000).27. M. F. Tsou, A. Hayashi, L. R. DeBella, G. McGrath, L. S.Rose, Development 129, 4469 (2002).

28. L. S. Rose, K. Kemphues, Development 125, 1337(1998).

29. S. W. Grill, J. Howard, E. Schaffer, E. H. K. Stelzer, A. A.Hyman, unpublished data.

30. We thank A. Ashford, J. Corbitt, T. K. Harden, M.Glozter, A. Hyman, M. Koelle, Y. Kohara, G. Seydoux,and V. Simanis for reagents; A. Debant and S. Schmidtfor advice; K. Baumer for technical support; M.Brauchle for movie S1; and K. Afshar, M. Delattre, andM. Labouesse for critical reading of the manuscript.Supported by Swiss National Science Foundationgrant 31-62102.00 (P.G.) and NIH grant GM62338(D.P.S.).

Supporting Online Materialwww.sciencemag.org/cgi/content/full/1084146/DC1Materials and MethodsFigs. S1 to S4ReferencesMovies S1 to S7

28 February 2003; accepted 8 May 2003Published online 15 May 2003;10.1126/science.1084146Include this information when citing this paper.

Transmission Dynamics of the EtiologicalAgent of SARS in Hong Kong: Impact of

Public Health InterventionsSteven Riley,1*† Christophe Fraser,1* Christl A. Donnelly,1

Azra C. Ghani,1 Laith J. Abu-Raddad,1 Anthony J. Hedley,2

Gabriel M. Leung,2 Lai-Ming Ho,2 Tai-Hing Lam,2

Thuan Q. Thach,2 Patsy Chau,2 King-Pan Chan,2 Su-Vui Lo,3

Pak-Yin Leung,4 Thomas Tsang,4 William Ho,5 Koon-Hung Lee,5

Edith M. C. Lau,6 Neil M. Ferguson,1 Roy M. Anderson1

We present an analysis of the first 10 weeks of the severe acute respiratorysyndrome (SARS) epidemic in Hong Kong. The epidemic to date has beencharacterized by two large clusters—initiated by two separate “super-spread”events (SSEs)—and by ongoing community transmission. By fitting a stochasticmodel to data on 1512 cases, including these clusters, we show that theetiological agent of SARS is moderately transmissible. Excluding SSEs, weestimate that 2.7 secondary infections were generated per case on average atthe start of the epidemic, with a substantial contribution from hospital trans-mission. Transmission rates fell during the epidemic, primarily as a result ofreductions in population contact rates and improved hospital infectioncontrol, but also because of more rapid hospital attendance by symptomaticindividuals. As a result, the epidemic is now in decline, although continuedvigilance is necessary for this to be maintained. Restrictions on longer rangepopulation movement are shown to be a potentially useful additional con-trol measure in some contexts. We estimate that most currently infectedpersons are now hospitalized, which highlights the importance of control ofnosocomial transmission.

The evolution and spread of the etiologicalagent of severe acute respiratory syndrome(SARS), a novel coronavirus (1), has resultedin an unparalleled international effort coordi-nated by the World Health Organization(WHO) to characterize the virus, develop

diagnostic tests, and formulate optimal treat-ment protocols to reduce morbidity and mor-tality (2–4). Great progress has been madewith, for example, the full sequence of theRNA virus reported on 13 April 2003 (5, 6).The epidemic apparently originated in early

November in the Guandong province of thePeople’s Republic of China, and then spreadrapidly throughout the world via air travel.As of 21 May 2003, 7956 cases have beenreported to WHO from 28 countries, with 666deaths recorded. The epidemics in HongKong, mainland China, Singapore, Taiwan,and Toronto (Canada) have been of partic-ular concern because of the multiple gen-erations of local transmission seen in thoseareas. The extent of these epidemics hasbeen worsened by the occurrence of largeclusters of infection linked to single indi-viduals and/or spatial locations.

The rate of spread of an epidemic—andwhether such spread is self-sustaining—depends on the magnitude of a key epidemi-ological parameter, the basic reproductionnumber (R0), defined as the average numberof secondary cases generated by one primarycase in a susceptible population (7). After theintroduction of an agent into a population, a

1Department of Infectious Disease Epidemiology, Fac-ulty of Medicine, Imperial College London, ExhibitionRoad, London SW7 2AZ, UK. 2Department of Com-munity Medicine, University of Hong Kong, 21 Sas-soon Road, Pokfulam, Hong Kong. 3Research Office,Health, Welfare and Food Bureau, Government of theHong Kong Special Administrative Region, 19th Floor,Murray Building, Garden Road, Hong Kong. 4Depart-ment of Health, Government of the Hong Kong Spe-cial Administrative Region, Wu Chung House, 213Queen’s Road East, Wanchai, Hong Kong. 5Hong KongHospital Authority, 147B Argyle Street, Kowloon,Hong Kong. 6Department of Community and FamilyMedicine, Chinese University of Hong Kong, School ofPublic Health, Prince of Wales Hospital, Shatin, N.T.,Hong Kong.

*These authors contributed equally to this work.†To whom correspondence should be addressed. E-mail: [email protected]

R E P O R T S

www.sciencemag.org SCIENCE VOL 300 20 JUNE 2003 1961

the cell cortex is a rate-limiting step for forcegenerator function. Given that cortically lo-calized G!i and Pins also dictate spindlepositioning in Drosophila (10), we proposethat modulation of cortical G! signaling togenerate defined pulling forces on astral mi-crotubules is a conserved mechanism totranslate polarity cues into appropriate spin-dle positioning.

References and Notes1. J. A. Knoblich, Nature Rev. Mol. Cell Biol. 2, 11(2001).

2. P. Gonczy, Trends Cell Biol. 12, 332 (2002).3. J. Pellettieri, G. Seydoux, Science 298, 1946 (2002).4. K. J. Kemphues, J. R. Priess, D. G. Morton, N. Cheng,Cell 52, 311 (1988).

5. S. W. Grill, P. Gonczy, E. H. Stelzer, A. A. Hyman,Nature 409, 630 (2001).

6. M. Gotta, J. Ahringer, Nature Cell Biol. 3, 297 (2001).7. F. Yu, X. Morin, Y. Cai, X. Yang, W. Chia, Cell 100, 399(2000).

8. M. Schaefer, A. Shevchenko, J. A. Knoblich, Curr. Biol.10, 353 (2000).

9. M. Schaefer, M. Petronczki, D. Dorner, M. Forte, J. A.Knoblich, Cell 107, 183 (2001).

10. Y. Cai, F. Yu, S. Lin, W. Chia, X. Yang, Cell 112, 51(2003).

11. P. Gonczy et al., Nature 408, 331 (2000).12. Examination of the genome sequence and sequencingof full-length cDNAs revealed that gpr-1 (WormBasecode F22B7.13) and gpr-2 (C38C10.4) are both ex-pressed and that they result from a recent geneduplication event. We found a single gpr homolog inthe available C. briggsae genome (CBG10100), whoseinactivation caused a phenotype indistinguishablefrom that observed in C. elegans.

13. B. Zhang et al., Nature 390, 477 (1997).14. K. Colombo, P. Gonczy, unpublished data.15. B. Etemad-Moghadam, S. Guo, K. J. Kemphues, Cell83, 743 (1995).

16. L. Boyd, S. Guo, D. Levitan, D. T. Stinchcomb, K. J.Kemphues, Development 122, 3075 (1996).

17. S. Guo, K. J. Kemphues, Cell 81, 611 (1995).18. T. J. Hung, K. J. Kemphues, Development 126, 127(1999).

19. S. Strome, W. B. Wood, Cell 35, 15 (1983).20. K. J. Reese, M. A. Dunn, J. A. Waddle, G. Seydoux, Mol.Cell 6, 445 (2000).

21. L. De Vries et al., Proc. Natl. Acad. Sci. U.S.A. 97,14364 (2000).

22. R. J. Kimple et al., J. Biol. Chem. 276, 29275 (2001).23. M. Natochin, K. G. Gasimov, N. O. Artemyev, Bio-chemistry 40, 5322 (2001).

24. See supporting online material.25. Average peak velocities in "m/s (#SEM at the 0.95confidence interval) were as follows (A, anterior spin-dle pole; P, posterior spindle pole): gpr-1/2(RNAi), A:0.20 # 0.03, P: 0.18 # 0.04, n $ 15; goa-1/gpa-

16(RNAi), A: 0.21 # 0.04, P: 0.22 # 0.03, n $ 15;par-3(it71) gpr-1/2(RNAi), A: 0.18 # 0.04, P: 0.18 #0.03, n $ 13. Values for par-3(it71) were A: 0.89 #0.11, P: 0.90 # 0.07, n $ 20 (5).

26. K. G. Miller, J. B. Rand, Genetics 156, 1649 (2000).27. M. F. Tsou, A. Hayashi, L. R. DeBella, G. McGrath, L. S.Rose, Development 129, 4469 (2002).

28. L. S. Rose, K. Kemphues, Development 125, 1337(1998).

29. S. W. Grill, J. Howard, E. Schaffer, E. H. K. Stelzer, A. A.Hyman, unpublished data.

30. We thank A. Ashford, J. Corbitt, T. K. Harden, M.Glozter, A. Hyman, M. Koelle, Y. Kohara, G. Seydoux,and V. Simanis for reagents; A. Debant and S. Schmidtfor advice; K. Baumer for technical support; M.Brauchle for movie S1; and K. Afshar, M. Delattre, andM. Labouesse for critical reading of the manuscript.Supported by Swiss National Science Foundationgrant 31-62102.00 (P.G.) and NIH grant GM62338(D.P.S.).

Supporting Online Materialwww.sciencemag.org/cgi/content/full/1084146/DC1Materials and MethodsFigs. S1 to S4ReferencesMovies S1 to S7

28 February 2003; accepted 8 May 2003Published online 15 May 2003;10.1126/science.1084146Include this information when citing this paper.

Transmission Dynamics of the EtiologicalAgent of SARS in Hong Kong: Impact of

Public Health InterventionsSteven Riley,1*† Christophe Fraser,1* Christl A. Donnelly,1

Azra C. Ghani,1 Laith J. Abu-Raddad,1 Anthony J. Hedley,2

Gabriel M. Leung,2 Lai-Ming Ho,2 Tai-Hing Lam,2

Thuan Q. Thach,2 Patsy Chau,2 King-Pan Chan,2 Su-Vui Lo,3

Pak-Yin Leung,4 Thomas Tsang,4 William Ho,5 Koon-Hung Lee,5

Edith M. C. Lau,6 Neil M. Ferguson,1 Roy M. Anderson1

We present an analysis of the first 10 weeks of the severe acute respiratorysyndrome (SARS) epidemic in Hong Kong. The epidemic to date has beencharacterized by two large clusters—initiated by two separate “super-spread”events (SSEs)—and by ongoing community transmission. By fitting a stochasticmodel to data on 1512 cases, including these clusters, we show that theetiological agent of SARS is moderately transmissible. Excluding SSEs, weestimate that 2.7 secondary infections were generated per case on average atthe start of the epidemic, with a substantial contribution from hospital trans-mission. Transmission rates fell during the epidemic, primarily as a result ofreductions in population contact rates and improved hospital infectioncontrol, but also because of more rapid hospital attendance by symptomaticindividuals. As a result, the epidemic is now in decline, although continuedvigilance is necessary for this to be maintained. Restrictions on longer rangepopulation movement are shown to be a potentially useful additional con-trol measure in some contexts. We estimate that most currently infectedpersons are now hospitalized, which highlights the importance of control ofnosocomial transmission.

The evolution and spread of the etiologicalagent of severe acute respiratory syndrome(SARS), a novel coronavirus (1), has resultedin an unparalleled international effort coordi-nated by the World Health Organization(WHO) to characterize the virus, develop

diagnostic tests, and formulate optimal treat-ment protocols to reduce morbidity and mor-tality (2–4). Great progress has been madewith, for example, the full sequence of theRNA virus reported on 13 April 2003 (5, 6).The epidemic apparently originated in early

November in the Guandong province of thePeople’s Republic of China, and then spreadrapidly throughout the world via air travel.As of 21 May 2003, 7956 cases have beenreported to WHO from 28 countries, with 666deaths recorded. The epidemics in HongKong, mainland China, Singapore, Taiwan,and Toronto (Canada) have been of partic-ular concern because of the multiple gen-erations of local transmission seen in thoseareas. The extent of these epidemics hasbeen worsened by the occurrence of largeclusters of infection linked to single indi-viduals and/or spatial locations.

The rate of spread of an epidemic—andwhether such spread is self-sustaining—depends on the magnitude of a key epidemi-ological parameter, the basic reproductionnumber (R0), defined as the average numberof secondary cases generated by one primarycase in a susceptible population (7). After theintroduction of an agent into a population, a

1Department of Infectious Disease Epidemiology, Fac-ulty of Medicine, Imperial College London, ExhibitionRoad, London SW7 2AZ, UK. 2Department of Com-munity Medicine, University of Hong Kong, 21 Sas-soon Road, Pokfulam, Hong Kong. 3Research Office,Health, Welfare and Food Bureau, Government of theHong Kong Special Administrative Region, 19th Floor,Murray Building, Garden Road, Hong Kong. 4Depart-ment of Health, Government of the Hong Kong Spe-cial Administrative Region, Wu Chung House, 213Queen’s Road East, Wanchai, Hong Kong. 5Hong KongHospital Authority, 147B Argyle Street, Kowloon,Hong Kong. 6Department of Community and FamilyMedicine, Chinese University of Hong Kong, School ofPublic Health, Prince of Wales Hospital, Shatin, N.T.,Hong Kong.

*These authors contributed equally to this work.†To whom correspondence should be addressed. E-mail: [email protected]

R E P O R T S

www.sciencemag.org SCIENCE VOL 300 20 JUNE 2003 1961

Pandemic Potential of a Strain ofInfluenza A (H1N1): Early Findings

Christophe Fraser,1* Christl A. Donnelly,1* Simon Cauchemez,1 William P. Hanage,1Maria D. Van Kerkhove,1 T. Déirdre Hollingsworth,1 Jamie Griffin,1 Rebecca F. Baggaley,1Helen E. Jenkins,1 Emily J. Lyons,1 Thibaut Jombart,1 Wes R. Hinsley,1 Nicholas C. Grassly,1

Francois Balloux,1 Azra C. Ghani,1 Neil M. Ferguson1†;

Andrew Rambaut,2 Oliver G. Pybus3;

Hugo Lopez-Gatell,4 Celia M. Alpuche-Aranda,5 Ietza Bojorquez Chapela,4 Ethel Palacios Zavala4;

Dulce Ma. Espejo Guevara6;

Francesco Checchi,7 Erika Garcia,7 Stephane Hugonnet,7 Cathy Roth7

The WHO Rapid Pandemic Assessment Collaboration‡A novel influenza A (H1N1) virus has spread rapidly across the globe. Judging its pandemicpotential is difficult with limited data, but nevertheless essential to inform appropriate healthresponses. By analyzing the outbreak in Mexico, early data on international spread, and viralgenetic diversity, we make an early assessment of transmissibility and severity. Our estimatessuggest that 23,000 (range 6000 to 32,000) individuals had been infected in Mexico by late April,giving an estimated case fatality ratio (CFR) of 0.4% (range: 0.3 to 1.8%) based on confirmed andsuspected deaths reported to that time. In a community outbreak in the small community of LaGloria, Veracruz, no deaths were attributed to infection, giving an upper 95% bound on CFR of0.6%. Thus, although substantial uncertainty remains, clinical severity appears less than that seenin the 1918 influenza pandemic but comparable with that seen in the 1957 pandemic. Clinicalattack rates in children in La Gloria were twice that in adults (<15 years of age: 61%; ≥15 years:29%). Three different epidemiological analyses gave basic reproduction number (R0) estimates inthe range of 1.4 to 1.6, whereas a genetic analysis gave a central estimate of 1.2. This range ofvalues is consistent with 14 to 73 generations of human-to-human transmission having occurred inMexico to late April. Transmissibility is therefore substantially higher than that of seasonal flu, andcomparable with lower estimates of R0 obtained from previous influenza pandemics.

On29April 2009, the World Health Orga-nization (WHO) announced that the rapidglobal spread of a strain of influenza A

(H1N1) virus detected in the previous week war-ranted moving the global pandemic alert level tophase 5 (www.who.int/csr/disease/swineflu/).Phase 5 indicates sustained human-to-human trans-mission of a novel influenza strain of animal originin one WHO region of the world, and exportedcases detected in other regions. In this outbreak,the earliest affected countrymayhave beenMexico,with many cases in other nations associated with

travels from that country. There are uncertaintiesabout all aspects of this outbreak, including thevirulence, transmissibility, and origin of the virus,and this in turn results in uncertainty in judgingthe pandemic potential of the virus and when re-active public health responses, such as recommen-dations to stay at home or to close schools, shouldbe implemented in individual countries. Here wereport findings of key early investigations into theoutbreak that could aid such policy decisions.

The presence of fatalities [29 confirmed plus88 suspected deaths in Mexico as of 4 May 2009(1), 1 confirmed in the United States as of 5 May2009 (2)] is not necessarily indicative of the viru-lence of the infection. The interpretation of thesestatistics depends on the total number of infec-tions, including those with mild infection or whoare asymptomatic, which is currently unknown,given the absence of a specific serological test forthe new H1N1 influenza strain and associatedpopulation-level screening. As of 4 May 2009,11,356 suspected and 822 laboratory-confirmedcases have been reported inMexico (1), but thesemay represent an underestimate of true case num-bers as surveillance has understandably focused onsevere cases. Furthermore, severe cases in olderindividuals will be more difficult to identify be-cause of the higher rate of respiratory illness in

those over 60 years of age (3), and this couldresult in an underestimate of overall morbidity.Right censoring of mortality data, which occurswhen additional deaths subsequently arise amongcases already included in surveillance data, canalso bias estimates of the true case fatality ratio(4). Finally, suspected deaths may not all havebeen caused by infection with the novel virus.These uncertainties necessarily affect any estimateof the case fatality ratio (CFR).

On the basis of international travel patterns,we would expect a proportion of cases of anyinfection spreading widely in Mexico to be ex-ported by travelers (5). Owing to intense surveil-lance for influenza-like illness in those returningfromMexico, ascertainment of early cases in new-ly affected countries was almost certainly morecomplete and rapid than local surveillance of mildcases in Mexico. Airline passenger flow out ofMexico shows a significant correlation with thefrequency of detected confirmed cases worldwide(Spearman correlation coefficient: 0.56, P= 0.004)(Fig. 1,A andB).We thus use data on cases amongtravelers and backcalculationmethods to estimatethe total number of people infected in Mexico.Key underlying assumptions in this analysis arethat population mixing in Mexico is equally like-ly between Mexican residents and tourists, andtourists and Mexican residents are at equal riskof infection (despite demographic and other dif-ferences). If infections are concentrated away fromtraveler destinations (Fig. 1E presents the spatialdistribution, by state, of cases within Mexico by5May), the number of people infected inMexicowill be underestimated, and conversely will beoverestimated if the epidemic has dispropor-tionately affected geographical zones visited bytravelers. Under the assumption that reporting ofinfections in travelers was complete, we estimatedthe number of infections that occurred in Mexicoby lateApril from amodel of the interval-censoredcountry case counts, which varied between 18,000and 32,000 (Table 1), depending on the meanduration of stay of tourists assumed, with perhapsthe most credible single value (based on journeyduration data) being 23,000. An alternative modelthat assumed at least one case had been confirmedin every country affected by late April gave lowerestimates of the number infected inMexico, in therange of 6000 to 11,000. However, this modelmay be viewed as a worst case (from the perspec-tive of resulting CFR estimates), and it fitted theobserved number of exported cases in key coun-tries (such as the United States and Canada) sub-stantially worse than did the first model. We used30 April 2009 as the cut-off date for the dataanalyzed, but the case data analyzed are subjectto delays (clinical onset, testing, and reporting) ofup to 1 week, so these estimates may be more rep-resentative of infections up to 23 April. The epi-demic has subsequently spread further, althoughthe impact of the nonpharmaceutical interven-tions introduced in Mexico is not yet known.

On the basis of the 9 confirmed and 92 sus-pected deaths that were reported by 30 April 2009

1MRC Centre for Outbreak Analysis and Modelling, Departmentof Infectious Disease Epidemiology, Imperial College London,Faculty of Medicine, Norfolk Place, London W2 1PG, UK.2Institute of Evolutionary Biology, University of Edinburgh,Ashworth Laboratories, Edinburgh EH9 3JT, UK. 3Departmentof Zoology, University of Oxford, South Parks Road, OxfordOX1 3PS, UK. 4Directorate General of Epidemiology, FCO.De P. Miranda, 177 5th Floor, Mexico City, 01480, Mexico.5National Institute of Epidemiological Diagnosis and Reference,Prolongación Carpio No. 470 (3° piso), Col Santo Tomás,México City, C.P. 11340, Mexico. 6Secretaría de Salud -Servicios de Salud de Veracruz Soconusco No. 36, ColoniaAguacatal, C.P. 910 Xalapa, Veracruz, México State. 7WorldHealth Organization.

*These authors contributed equally to this work.†To whom correspondence should be addressed. E-mail:[email protected]‡All authors are members of this collaboration.

www.sciencemag.org SCIENCE VOL 324 19 JUNE 2009 1557

REPORTS

on

Oct

ober

5, 2

009

www.

scie

ncem

ag.o

rgDo

wnlo

aded

from

Pandemic Potential of a Strain ofInfluenza A (H1N1): Early Findings

Christophe Fraser,1* Christl A. Donnelly,1* Simon Cauchemez,1 William P. Hanage,1Maria D. Van Kerkhove,1 T. Déirdre Hollingsworth,1 Jamie Griffin,1 Rebecca F. Baggaley,1Helen E. Jenkins,1 Emily J. Lyons,1 Thibaut Jombart,1 Wes R. Hinsley,1 Nicholas C. Grassly,1

Francois Balloux,1 Azra C. Ghani,1 Neil M. Ferguson1†;

Andrew Rambaut,2 Oliver G. Pybus3;

Hugo Lopez-Gatell,4 Celia M. Alpuche-Aranda,5 Ietza Bojorquez Chapela,4 Ethel Palacios Zavala4;

Dulce Ma. Espejo Guevara6;

Francesco Checchi,7 Erika Garcia,7 Stephane Hugonnet,7 Cathy Roth7

The WHO Rapid Pandemic Assessment Collaboration‡A novel influenza A (H1N1) virus has spread rapidly across the globe. Judging its pandemicpotential is difficult with limited data, but nevertheless essential to inform appropriate healthresponses. By analyzing the outbreak in Mexico, early data on international spread, and viralgenetic diversity, we make an early assessment of transmissibility and severity. Our estimatessuggest that 23,000 (range 6000 to 32,000) individuals had been infected in Mexico by late April,giving an estimated case fatality ratio (CFR) of 0.4% (range: 0.3 to 1.8%) based on confirmed andsuspected deaths reported to that time. In a community outbreak in the small community of LaGloria, Veracruz, no deaths were attributed to infection, giving an upper 95% bound on CFR of0.6%. Thus, although substantial uncertainty remains, clinical severity appears less than that seenin the 1918 influenza pandemic but comparable with that seen in the 1957 pandemic. Clinicalattack rates in children in La Gloria were twice that in adults (<15 years of age: 61%; ≥15 years:29%). Three different epidemiological analyses gave basic reproduction number (R0) estimates inthe range of 1.4 to 1.6, whereas a genetic analysis gave a central estimate of 1.2. This range ofvalues is consistent with 14 to 73 generations of human-to-human transmission having occurred inMexico to late April. Transmissibility is therefore substantially higher than that of seasonal flu, andcomparable with lower estimates of R0 obtained from previous influenza pandemics.

On29April 2009, the World Health Orga-nization (WHO) announced that the rapidglobal spread of a strain of influenza A

(H1N1) virus detected in the previous week war-ranted moving the global pandemic alert level tophase 5 (www.who.int/csr/disease/swineflu/).Phase 5 indicates sustained human-to-human trans-mission of a novel influenza strain of animal originin one WHO region of the world, and exportedcases detected in other regions. In this outbreak,the earliest affected countrymayhave beenMexico,with many cases in other nations associated with

travels from that country. There are uncertaintiesabout all aspects of this outbreak, including thevirulence, transmissibility, and origin of the virus,and this in turn results in uncertainty in judgingthe pandemic potential of the virus and when re-active public health responses, such as recommen-dations to stay at home or to close schools, shouldbe implemented in individual countries. Here wereport findings of key early investigations into theoutbreak that could aid such policy decisions.

The presence of fatalities [29 confirmed plus88 suspected deaths in Mexico as of 4 May 2009(1), 1 confirmed in the United States as of 5 May2009 (2)] is not necessarily indicative of the viru-lence of the infection. The interpretation of thesestatistics depends on the total number of infec-tions, including those with mild infection or whoare asymptomatic, which is currently unknown,given the absence of a specific serological test forthe new H1N1 influenza strain and associatedpopulation-level screening. As of 4 May 2009,11,356 suspected and 822 laboratory-confirmedcases have been reported inMexico (1), but thesemay represent an underestimate of true case num-bers as surveillance has understandably focused onsevere cases. Furthermore, severe cases in olderindividuals will be more difficult to identify be-cause of the higher rate of respiratory illness in

those over 60 years of age (3), and this couldresult in an underestimate of overall morbidity.Right censoring of mortality data, which occurswhen additional deaths subsequently arise amongcases already included in surveillance data, canalso bias estimates of the true case fatality ratio(4). Finally, suspected deaths may not all havebeen caused by infection with the novel virus.These uncertainties necessarily affect any estimateof the case fatality ratio (CFR).

On the basis of international travel patterns,we would expect a proportion of cases of anyinfection spreading widely in Mexico to be ex-ported by travelers (5). Owing to intense surveil-lance for influenza-like illness in those returningfromMexico, ascertainment of early cases in new-ly affected countries was almost certainly morecomplete and rapid than local surveillance of mildcases in Mexico. Airline passenger flow out ofMexico shows a significant correlation with thefrequency of detected confirmed cases worldwide(Spearman correlation coefficient: 0.56, P= 0.004)(Fig. 1,A andB).We thus use data on cases amongtravelers and backcalculationmethods to estimatethe total number of people infected in Mexico.Key underlying assumptions in this analysis arethat population mixing in Mexico is equally like-ly between Mexican residents and tourists, andtourists and Mexican residents are at equal riskof infection (despite demographic and other dif-ferences). If infections are concentrated away fromtraveler destinations (Fig. 1E presents the spatialdistribution, by state, of cases within Mexico by5May), the number of people infected inMexicowill be underestimated, and conversely will beoverestimated if the epidemic has dispropor-tionately affected geographical zones visited bytravelers. Under the assumption that reporting ofinfections in travelers was complete, we estimatedthe number of infections that occurred in Mexicoby lateApril from amodel of the interval-censoredcountry case counts, which varied between 18,000and 32,000 (Table 1), depending on the meanduration of stay of tourists assumed, with perhapsthe most credible single value (based on journeyduration data) being 23,000. An alternative modelthat assumed at least one case had been confirmedin every country affected by late April gave lowerestimates of the number infected inMexico, in therange of 6000 to 11,000. However, this modelmay be viewed as a worst case (from the perspec-tive of resulting CFR estimates), and it fitted theobserved number of exported cases in key coun-tries (such as the United States and Canada) sub-stantially worse than did the first model. We used30 April 2009 as the cut-off date for the dataanalyzed, but the case data analyzed are subjectto delays (clinical onset, testing, and reporting) ofup to 1 week, so these estimates may be more rep-resentative of infections up to 23 April. The epi-demic has subsequently spread further, althoughthe impact of the nonpharmaceutical interven-tions introduced in Mexico is not yet known.

On the basis of the 9 confirmed and 92 sus-pected deaths that were reported by 30 April 2009

1MRC Centre for Outbreak Analysis and Modelling, Departmentof Infectious Disease Epidemiology, Imperial College London,Faculty of Medicine, Norfolk Place, London W2 1PG, UK.2Institute of Evolutionary Biology, University of Edinburgh,Ashworth Laboratories, Edinburgh EH9 3JT, UK. 3Departmentof Zoology, University of Oxford, South Parks Road, OxfordOX1 3PS, UK. 4Directorate General of Epidemiology, FCO.De P. Miranda, 177 5th Floor, Mexico City, 01480, Mexico.5National Institute of Epidemiological Diagnosis and Reference,Prolongación Carpio No. 470 (3° piso), Col Santo Tomás,México City, C.P. 11340, Mexico. 6Secretaría de Salud -Servicios de Salud de Veracruz Soconusco No. 36, ColoniaAguacatal, C.P. 910 Xalapa, Veracruz, México State. 7WorldHealth Organization.

*These authors contributed equally to this work.†To whom correspondence should be addressed. E-mail:[email protected]‡All authors are members of this collaboration.

www.sciencemag.org SCIENCE VOL 324 19 JUNE 2009 1557

REPORTS

on

Octo

ber 5

, 200

9 ww

w.sc

ience

mag

.org

Down

loade

d fro

m

(6) and assuming similar times from infection toconfirmation and from infection to death, we esti-mated CFRs in the range of 0.3 to 0.6% from theinterval-censored case count model, based on con-firmed and suspected deaths combined, or 0.03 to0.05% for confirmed deaths only. Using the alter-native, more pessimistic, country presence/absencemodel, we estimatedCFRs of 0.9 to 1.8%based onsuspected and confirmed deaths, and 0.08 to0.16% from the confirmed deaths alone. Theseestimates have already changed somewhat as aresult of data available after 30 April, but we de-liberately report the earlier analysis because itformed part of the evidence base used by WHOto move to phase 5.

Another source of information on severitycomes from the large outbreak of respiratory dis-ease seen in the small, isolated community of LaGloria in Veracruz province, one case of whichhas been confirmed to have been caused by thenovel H1N1 strain. It is possible that other vi-ruses were circulating at the same time as theoutbreak, but the overall attack rate is substan-tially larger than would be expected for a sea-sonal influenza outbreak. No fatalities among 616cases have been attributed to infection during thefull period of surveillance of that outbreak (Fig.3A), giving a 95% confidence interval (CI) of 0to 0.60%.

Data on themagnitude of the current outbreakin Mexico can also be used to estimate the trans-missibility of the virus if the start date of theoutbreak is known or can be estimated. Epide-miological investigations into the emergence ofthe virus inMexico have focused on the La Gloriaoutbreak, where the first case in that outbreak isthought to have occurred around 15 February 2009(Fig. 3A).

An alternative approach to estimating the startdate of the outbreak is to look at the diversity inthe genetic sequences of viral samples collectedfrom confirmed cases, assuming that diversityaccumulates according to amolecular clockmodel.Twenty-three complete publicly available hemag-glutinin (HA) gene sequences from cases notlinked in epidemiological clusters were analyzedwith a Bayesian coalescent method that assumesexponential growth of the viral population (7).This yielded an estimate of the time of mostrecent common ancestor (TMRCA) of 12January 2009 [95% credible interval (CrI): 3November 2008 to 2 March 2009]. The geneticmodel also gave an estimate of the doublingtime of the epidemic of 10 days (95% CrI: 4.5to 37.5 days) (Fig. 2). Assuming exponentialgrowth, the TMRCA is a reasonable estimateof the start of the outbreak, although it is for-mally an upper bound due to incomplete sam-pling of the epidemic and the effects of theexponential model prior to distribution. These find-ings from a population genetic analysis are con-sistent with the epidemiological investigation ofboth the start and magnitude of the current epi-demic in Mexico. Figure 2 also shows a pre-liminary version of this analysis based on the first

11 sequences, which gave similar estimates high-lighting the power of these methods. [See (8) forfurther sensitivity analysis and methods.]

The reproduction number, defined as the num-ber of cases one case generates on average overthe course of their infectious period, is a key mea-sure of transmissibility and can be estimated in anumber of ways from the data currently available.

First, by assuming exponential growth, thegrowth rate of the epidemic (r) can be inferredfrom estimates of the current cumulative numberof infections (Yf) and estimated start date and sizefor the outbreak (t0 and Y0, respectively). Thebasic reproduction number (R0) can be estimatedfrom the exponential growth rate if one alsoassumes that the generation time distribution for

the new H1N1 strain is similar to that of otherstrains of seasonal and pandemic viruses (9, 10)[Table 1 and (8)]. Using the date of 15 Februaryas the first case of the La Gloria outbreak (8)gives reproduction number estimates of between1.31 and 1.42, depending on which variant ofthe geographical backcalculation model is used.Extending a more sophisticated Bayesian estima-tion method (11) that allows for stochastic varia-bility intrinsic to epidemic dynamics and parameteruncertainty gave similar but slightly higher esti-mates for R0 with wider ranges: posterior median =1.40; 95% CrI: 1.15 to 1.90 (Fig. 1C).

Second, by assuming a prior distribution onthe generation time distribution informed by pre-vious estimates of influenza, the Bayesian coa-

A

B

C

D E

Fig. 1. (A). The number of passengers flying out of Mexico by actual destination and the number ofconfirmed cases as reported on 30 April 2009. (B) The number of cases exported to country j asreported on 30 April 2009 as a function of the estimated average number of foreign travelers inMexico from country j on any given day in March or April. Black circles: minimal number based onone exposure per epidemiological cluster; filled red circles, total number of confirmed cases. (C)Mean assumed generation time distribution (red) and 100 illustrative draws from the priordistribution, and (D) corresponding posterior distribution of R0 estimates for a stochastic model ofan epidemic within Mexico with travelers infected at a rate proportional to the estimated density oftravelers per local resident. The two bar charts correspond to a 7-day delay between infection andconfirmation (blue) and no delay (orange) in cases among travelers. (E) Number of acute respiratoryinfection cases per 100,000 inhabitants by state as reported on 5 May 2009 (1), demonstrating spatialdistribution of disease within Mexico.

19 JUNE 2009 VOL 324 SCIENCE www.sciencemag.org1558

REPORTS

on

Oct

ober

5, 2

009

www.

scie

ncem

ag.o

rgDo

wnlo

aded

from

50 www.thelancet.com/infection Vol 14 January 2014

Articles

Middle East respiratory syndrome coronavirus: quantifi cation of the extent of the epidemic, surveillance biases, and transmissibilitySimon Cauchemez*, Christophe Fraser*, Maria D Van Kerkhove, Christl A Donnelly, Steven Riley, Andrew Rambaut, Vincent Enouf, Sylvie van der Werf, Neil M Ferguson

SummaryBackground The novel Middle East respiratory syndrome coronavirus (MERS-CoV) had, as of Aug 8, 2013, caused 111 virologically confi rmed or probable human cases of infection worldwide. We analysed epidemiological and genetic data to assess the extent of human infection, the performance of case detection, and the transmission potential of MERS-CoV with and without control measures.

Methods We assembled a comprehensive database of all confi rmed and probable cases from public sources and estimated the incubation period and generation time from case cluster data. Using data of numbers of visitors to the Middle East and their duration of stay, we estimated the number of symptomatic cases in the Middle East. We did independent analyses, looking at the growth in incident clusters, the growth in viral population, the reproduction number of cluster index cases, and cluster sizes to characterise the dynamical properties of the epidemic and the transmission scenario.

Findings The estimated number of symptomatic cases up to Aug 8, 2013, is 940 (95% CI 290–2200), indicating that at least 62% of human symptomatic cases have not been detected. We fi nd that the case-fatality ratio of primary cases detected via routine surveillance (74%; 95% CI 49–91) is biased upwards because of detection bias; the case-fatality ratio of secondary cases was 20% (7–42). Detection of milder cases (or clinical management) seemed to have improved in recent months. Analysis of human clusters indicated that chains of transmission were not self-sustaining when infection control was implemented, but that R in the absence of controls was in the range 0·8–1·3. Three independent data sources provide evidence that R cannot be much above 1, with an upper bound of 1·2–1·5.

Interpretation By showing that a slowly growing epidemic is underway either in human beings or in an animal reservoir, quantifi cation of uncertainty in transmissibility estimates, and provision of the fi rst estimates of the scale of the epidemic and extent of case detection biases, we provide valuable information for more informed risk assessment.

Funding Medical Research Council, Bill & Melinda Gates Foundation, EU FP7, and National Institute of General Medical Sciences.

IntroductionThe earliest known human infections with Middle East respiratory syndrome coronavirus (MERS-CoV) occurred in Jordan in March, 2012,1 with isolation and identifi cation of the virus from a patient in Saudi Arabia occurring some months later.2 By Aug 8, 2013, 94 virologically confi rmed human cases and 17 probable cases1,3–5 had been reported. Zoonotic exposure is suspected as the source of human infection, in view of the initially sporadic occurrence of cases together with the genetic similarity of MERS-CoV to bat coronaviruses.6 However, epidemiological investigations of cases did not fi nd a consistent pattern of exposure to animals or the environment,7 and, as of Aug 8, 2013, the virus had not been isolated from any animal species. Recently, however, camels from Oman and the Spanish Canary Islands were discovered to have been previously infected by the MERS-CoV or a closely related virus.8 The high apparent case-fatality ratio in reported cases is cause for concern: as of Aug 8, 2013, WHO reported that 46 of the 94 virologically

confi rmed cases had died;7 this death toll is expected to rise since some patients are still in hospital.9

Although progress has been made in characterising the epidemiology of MERS-CoV, many uncertainties remain.10 Little is known about the extent of human infection or the degree of detection bias towards more severe cases. If the severe cases currently being detected represent only a small sentinel minority of a much larger number of milder cases (as occurred early in the 2009 H1N1 pandemic in Mexico11), the case-fatality ratio might be substantially lower than what current surveillance data suggest. Conversely, for the severe acute respiratory syndrome (SARS) epidemic of 2003, little evidence of undetected mild or subclinical infections existed,12 even after detailed serological follow-up studies.13 In the absence of robust community-based serological surveys for MERS-CoV, assessment of the extent of undetected infection must rely on epidemiological inference. Another essential aspect of risk assessment is the characterisation of transmissibility, and a study14

Lancet Infect Dis 2014; 14: 50–56

Published OnlineNovember 13, 2013

http://dx.doi.org/10.1016/S1473-3099(13)70304-9

See Comment page 6

Copyright © Cauchemez et al. Open Access article distributed

under the terms of CC BY

*These authors contributed equally to this study

MRC Centre for Outbreak Analysis and Modelling,

Department of Infectious Disease Epidemiology, Imperial

College London, London, UK (S Cauchemez PhD,

Prof C Fraser PhD, MD Van Kerkhove PhD,

Prof CA Donnelly ScD, S Riley PhD,

Prof N M Ferguson DPhil); Institute of Evolutionary

Biology, Ashworth Laboratories, University of Edinburgh, Edinburgh, UK (Prof A Rambaut PhD); and

Institut Pasteur, Unit of Molecular Genetics of RNA

Viruses, UMR3569 CNRS, Université Paris Diderot

Sorbonne Paris Cité, Paris, France (V Enouf PhD,

Prof S van der Werf PhD)

Correspondence to:Prof Neil M Ferguson, MRC

Centre for Outbreak Analysis and Modelling, Department of

Infectious Disease Epidemiology, Imperial College London, Norfolk

Place, London W2 1PG, [email protected]

50 www.thelancet.com/infection Vol 14 January 2014

Articles

Middle East respiratory syndrome coronavirus: quantifi cation of the extent of the epidemic, surveillance biases, and transmissibilitySimon Cauchemez*, Christophe Fraser*, Maria D Van Kerkhove, Christl A Donnelly, Steven Riley, Andrew Rambaut, Vincent Enouf, Sylvie van der Werf, Neil M Ferguson

SummaryBackground The novel Middle East respiratory syndrome coronavirus (MERS-CoV) had, as of Aug 8, 2013, caused 111 virologically confi rmed or probable human cases of infection worldwide. We analysed epidemiological and genetic data to assess the extent of human infection, the performance of case detection, and the transmission potential of MERS-CoV with and without control measures.

Methods We assembled a comprehensive database of all confi rmed and probable cases from public sources and estimated the incubation period and generation time from case cluster data. Using data of numbers of visitors to the Middle East and their duration of stay, we estimated the number of symptomatic cases in the Middle East. We did independent analyses, looking at the growth in incident clusters, the growth in viral population, the reproduction number of cluster index cases, and cluster sizes to characterise the dynamical properties of the epidemic and the transmission scenario.

Findings The estimated number of symptomatic cases up to Aug 8, 2013, is 940 (95% CI 290–2200), indicating that at least 62% of human symptomatic cases have not been detected. We fi nd that the case-fatality ratio of primary cases detected via routine surveillance (74%; 95% CI 49–91) is biased upwards because of detection bias; the case-fatality ratio of secondary cases was 20% (7–42). Detection of milder cases (or clinical management) seemed to have improved in recent months. Analysis of human clusters indicated that chains of transmission were not self-sustaining when infection control was implemented, but that R in the absence of controls was in the range 0·8–1·3. Three independent data sources provide evidence that R cannot be much above 1, with an upper bound of 1·2–1·5.

Interpretation By showing that a slowly growing epidemic is underway either in human beings or in an animal reservoir, quantifi cation of uncertainty in transmissibility estimates, and provision of the fi rst estimates of the scale of the epidemic and extent of case detection biases, we provide valuable information for more informed risk assessment.

Funding Medical Research Council, Bill & Melinda Gates Foundation, EU FP7, and National Institute of General Medical Sciences.

IntroductionThe earliest known human infections with Middle East respiratory syndrome coronavirus (MERS-CoV) occurred in Jordan in March, 2012,1 with isolation and identifi cation of the virus from a patient in Saudi Arabia occurring some months later.2 By Aug 8, 2013, 94 virologically confi rmed human cases and 17 probable cases1,3–5 had been reported. Zoonotic exposure is suspected as the source of human infection, in view of the initially sporadic occurrence of cases together with the genetic similarity of MERS-CoV to bat coronaviruses.6 However, epidemiological investigations of cases did not fi nd a consistent pattern of exposure to animals or the environment,7 and, as of Aug 8, 2013, the virus had not been isolated from any animal species. Recently, however, camels from Oman and the Spanish Canary Islands were discovered to have been previously infected by the MERS-CoV or a closely related virus.8 The high apparent case-fatality ratio in reported cases is cause for concern: as of Aug 8, 2013, WHO reported that 46 of the 94 virologically

confi rmed cases had died;7 this death toll is expected to rise since some patients are still in hospital.9

Although progress has been made in characterising the epidemiology of MERS-CoV, many uncertainties remain.10 Little is known about the extent of human infection or the degree of detection bias towards more severe cases. If the severe cases currently being detected represent only a small sentinel minority of a much larger number of milder cases (as occurred early in the 2009 H1N1 pandemic in Mexico11), the case-fatality ratio might be substantially lower than what current surveillance data suggest. Conversely, for the severe acute respiratory syndrome (SARS) epidemic of 2003, little evidence of undetected mild or subclinical infections existed,12 even after detailed serological follow-up studies.13 In the absence of robust community-based serological surveys for MERS-CoV, assessment of the extent of undetected infection must rely on epidemiological inference. Another essential aspect of risk assessment is the characterisation of transmissibility, and a study14

Lancet Infect Dis 2014; 14: 50–56

Published OnlineNovember 13, 2013

http://dx.doi.org/10.1016/S1473-3099(13)70304-9

See Comment page 6

Copyright © Cauchemez et al. Open Access article distributed

under the terms of CC BY

*These authors contributed equally to this study

MRC Centre for Outbreak Analysis and Modelling,

Department of Infectious Disease Epidemiology, Imperial

College London, London, UK (S Cauchemez PhD,

Prof C Fraser PhD, MD Van Kerkhove PhD,

Prof CA Donnelly ScD, S Riley PhD,

Prof N M Ferguson DPhil); Institute of Evolutionary

Biology, Ashworth Laboratories, University of Edinburgh, Edinburgh, UK (Prof A Rambaut PhD); and

Institut Pasteur, Unit of Molecular Genetics of RNA

Viruses, UMR3569 CNRS, Université Paris Diderot

Sorbonne Paris Cité, Paris, France (V Enouf PhD,

Prof S van der Werf PhD)

Correspondence to:Prof Neil M Ferguson, MRC

Centre for Outbreak Analysis and Modelling, Department of

Infectious Disease Epidemiology, Imperial College London, Norfolk

Place, London W2 1PG, [email protected]

T h e n e w e ngl a nd j o u r na l o f m e dic i n e

n engl j med nejm.org 1

original article

Ebola Virus Disease in West Africa — The First 9 Months of the Epidemic

and Forward ProjectionsWHO Ebola Response Team*

Address reprint requests to Dr. Christl Donnelly at [email protected] or Dr. Christopher Dye at [email protected].

*The authors (members of the World Health Organization [WHO] Ebola Re-sponse team who contributed to this ar-ticle) are listed in the Appendix.

This article was published on September 23, 2014, at NEJM.org.

DOI: 10.1056/NEJMoa1411100Copyright © 2014 Massachusetts Medical Society.

A bs tr ac t

BackgroundOn March 23, 2014, the World Health Organization (WHO) was notified of an out-break of Ebola virus disease (EVD) in Guinea. On August 8, the WHO declared the epidemic to be a “public health emergency of international concern.”

MethodsBy September 14, 2014, a total of 4507 probable and confirmed cases, including 2296 deaths from EVD (Zaire species) had been reported from five countries in West Africa — Guinea, Liberia, Nigeria, Senegal, and Sierra Leone. We analyzed a detailed subset of data on 3343 confirmed and 667 probable Ebola cases collected in Guinea, Liberia, Nigeria, and Sierra Leone as of September 14.

ResultsThe majority of patients are 15 to 44 years of age (49.9% male), and we estimate that the case fatality rate is 70.8% (95% confidence interval [CI], 69 to 73) among per-sons with known clinical outcome of infection. The course of infection, including signs and symptoms, incubation period (11.4 days), and serial interval (15.3 days), is similar to that reported in previous outbreaks of EVD. On the basis of the initial periods of exponential growth, the estimated basic reproduction numbers (R0) are 1.71 (95% CI, 1.44 to 2.01) for Guinea, 1.83 (95% CI, 1.72 to 1.94) for Liberia, and 2.02 (95% CI, 1.79 to 2.26) for Sierra Leone. The estimated current reproduction numbers (R) are 1.81 (95% CI, 1.60 to 2.03) for Guinea, 1.51 (95% CI, 1.41 to 1.60) for Liberia, and 1.38 (95% CI, 1.27 to 1.51) for Sierra Leone; the corresponding doubling times are 15.7 days (95% CI, 12.9 to 20.3) for Guinea, 23.6 days (95% CI, 20.2 to 28.2) for Liberia, and 30.2 days (95% CI, 23.6 to 42.3) for Sierra Leone. As-suming no change in the control measures for this epidemic, by November 2, 2014, the cumulative reported numbers of confirmed and probable cases are predicted to be 5740 in Guinea, 9890 in Liberia, and 5000 in Sierra Leone, exceeding 20,000 in total.

ConclusionsThese data indicate that without drastic improvements in control measures, the numbers of cases of and deaths from EVD are expected to continue increasing from hundreds to thousands per week in the coming months.

The New England Journal of Medicine Downloaded from nejm.org on September 22, 2014. For personal use only. No other uses without permission.

Copyright © 2014 Massachusetts Medical Society. All rights reserved.

T h e n e w e ngl a nd j o u r na l o f m e dic i n e

n engl j med nejm.org 1

original article

Ebola Virus Disease in West Africa — The First 9 Months of the Epidemic

and Forward ProjectionsWHO Ebola Response Team*

Address reprint requests to Dr. Christl Donnelly at [email protected] or Dr. Christopher Dye at [email protected].

*The authors (members of the World Health Organization [WHO] Ebola Re-sponse team who contributed to this ar-ticle) are listed in the Appendix.

This article was published on September 23, 2014, at NEJM.org.

DOI: 10.1056/NEJMoa1411100Copyright © 2014 Massachusetts Medical Society.

A bs tr ac t

BackgroundOn March 23, 2014, the World Health Organization (WHO) was notified of an out-break of Ebola virus disease (EVD) in Guinea. On August 8, the WHO declared the epidemic to be a “public health emergency of international concern.”

MethodsBy September 14, 2014, a total of 4507 probable and confirmed cases, including 2296 deaths from EVD (Zaire species) had been reported from five countries in West Africa — Guinea, Liberia, Nigeria, Senegal, and Sierra Leone. We analyzed a detailed subset of data on 3343 confirmed and 667 probable Ebola cases collected in Guinea, Liberia, Nigeria, and Sierra Leone as of September 14.

ResultsThe majority of patients are 15 to 44 years of age (49.9% male), and we estimate that the case fatality rate is 70.8% (95% confidence interval [CI], 69 to 73) among per-sons with known clinical outcome of infection. The course of infection, including signs and symptoms, incubation period (11.4 days), and serial interval (15.3 days), is similar to that reported in previous outbreaks of EVD. On the basis of the initial periods of exponential growth, the estimated basic reproduction numbers (R0) are 1.71 (95% CI, 1.44 to 2.01) for Guinea, 1.83 (95% CI, 1.72 to 1.94) for Liberia, and 2.02 (95% CI, 1.79 to 2.26) for Sierra Leone. The estimated current reproduction numbers (R) are 1.81 (95% CI, 1.60 to 2.03) for Guinea, 1.51 (95% CI, 1.41 to 1.60) for Liberia, and 1.38 (95% CI, 1.27 to 1.51) for Sierra Leone; the corresponding doubling times are 15.7 days (95% CI, 12.9 to 20.3) for Guinea, 23.6 days (95% CI, 20.2 to 28.2) for Liberia, and 30.2 days (95% CI, 23.6 to 42.3) for Sierra Leone. As-suming no change in the control measures for this epidemic, by November 2, 2014, the cumulative reported numbers of confirmed and probable cases are predicted to be 5740 in Guinea, 9890 in Liberia, and 5000 in Sierra Leone, exceeding 20,000 in total.

ConclusionsThese data indicate that without drastic improvements in control measures, the numbers of cases of and deaths from EVD are expected to continue increasing from hundreds to thousands per week in the coming months.

The New England Journal of Medicine Downloaded from nejm.org on September 22, 2014. For personal use only. No other uses without permission.

Copyright © 2014 Massachusetts Medical Society. All rights reserved.

Page 3: Session 4: Modeling of potential pandemics

Lessons learnt• Epidemic response needs well-organised classical public health.

• Timeliness is most important.

• Diagnostics, prompt isolation, social distancing, risk communication.

• Require data systems, multidisciplinary cross-validation, sharing of data and samples.

• To date:

• Vaccines have arrived too late for outbreaks.

• Treatment as prevention is haphazard.

• Huge role for basic virological science, but why GOF with PPP?

Page 4: Session 4: Modeling of potential pandemics

Predicting pandemics

• H5N1, H7N9, MERS, Ebola clearly identified as threats prior to any GOF-PPP experiments.

• 2009 H1N1pdm, SARS - less so.

• Cause: surveillance gaps, not lack of understanding.

Page 5: Session 4: Modeling of potential pandemics

Supplementary Fig. 7). EA viruses showed the highest and most pro-longed virus shedding, closely followed by TRIG viruses; CS virusesshowed lower peak viral titres. Thus, the replicative advantage of EAviruses, together with the low prevalence of crossreactive antibodies toEA in swine (15% in 2000, 26% in 2004; Supplementary Table 4), mayhelp to explain the replacement of other SwIV lineages with EA viruses.

We tracked the evolution in our EA viruses of amino acids previ-ously associated with adaptation of avian influenza to other species15–17.Purported avian residues were maintained at most of these sites(Supplementary Table 5) despite the circulation of these viruses inswine for more than 30 years18,19. However, the PDZ- (post-synapticdensity protein, Drosophila disc large tumor suppressor and zonulaoccludens-1 protein) ligand at the 39 end of EA virus non-structural(NS) 1 genes showed significant host-specific evolution: early EuropeanEA viruses had the avian ESEV motif, with a change to GSEV/GPEV

motifs observed in several hosts. By 1999, most viruses sampled had theGPKV motif previously described from pigs16. CS and TRIG virusesthat contributed the NS gene to H1N1/2009 have a truncated NS gene,as do the antigenically variant Sw/HK/72/2007-like viruses. The role ofthe truncated NS gene in inter-species transmission clearly meritsfurther study. Furthermore, a modest but significant (P , 0.01) changein selection pressure was observed between European EA viruses iso-lated shortly after cross-species transmission (non-synonymous tosynonymous (dN/dS) substitution rate ratio of 0.24; 95% confidenceinterval 5 0.22–0.27) and those isolated later (dN/dS 5 0.17; 95% con-fidence interval 5 0.14–0.20), consistent with the hypothesis that host-specific selection increased viral adaptation after the introduction ofEA viruses into swine (Supplementary Table 6).

Our unique longitudinal study reveals a genetically and antigenicallydynamic SwIV population within a single region and provides a baselinefor future studies of the virus elsewhere. The epidemiology and evolu-tion of SwIV seem to be strongly shaped by gene flow among continentsand species, facilitating the reassortment of diverse lineages and occa-sionally resulting in antigenic change. Although we confirm that theH1N1/2009 virus was not generated within our study’s catchment, theprocesses of lineage emergence, importation, reassortment and replace-ment described here are probably representative of the H1N1/2009source population. We show that reassortments between EA andTRIG viruses do occur, generating reassortants that establish themselvesas stable lineages in swine. SwIV reassortants containing H1N1/2009-like genome segments have also been transiently detected5,20.

Despite clear evidence of inter-continental SwIV movement, geneflow is not so frequent that the global SwIV population acts as a singlegene pool (as observed for human influenza A21,22); instead a higherdiversity of mammalian-adapted viruses in global swine populations issupported. Crucially, the co-circulation of multiple SwIV lineages facili-tates the production of new genomic combinations. The evolutionaryconsequences of increased SwIV movement are hard to predict butrequire consideration given an increasingly globalized future.

a

b

Segment1 2 3 4 5 6 7 8

CSEATRIG

H1N1/2009Avian

Human H3N2

1974 1978 1982 1986 1990 1994 1998 2002 2006 2010Year

Figure 3 | Phylogenies and divergence times of the haemagglutinin genes ofclassical swine and European avian-like SwIV. a, CS; b, EA. Coloured boxesadjacent to tips show the lineage classification of each gene segment of SwIVisolated in this study. Arrows indicate the long branches that lead to newlydetected reassortant SwIV. Purple node bars represent 95% credible intervals oflineage divergence times. A fully detailed HA phylogeny including sequencenames is shown in Supplementary Fig. 3a.

4167/1999

1304/2003

1110/2006

Cal/4/2009

NS29/2009

1559/2008

Test antigens

CS H1N1 CS H1N2 TRIG H1N2 H1N1/09 EA H1N1 EA* H1N1

Sw/HK/4167/1999 CS H1N1 1:20,480 1:320 1:10,240 1:2,560 1:2,560 <1:10

Sw/HK/1304/2003 CS H1N2 1:1,280 1:2,560 1:640 1:80 1:40 <1:10

Sw/HK/1110/2006 TRIG H1N2 1:40,960 1:1,280 1:10,240 1:640 1:5,120 <1:10

Cal/04/2009 H1N1/2009 1:640 1:640 1:2,560 1:1,280 160 <1:10

Sw/HK/NS29/2009 EA H1N1 1:640 1:160 1:1,280 1:80 1:10,240 <1:10

Sw/HK/1559/2008 EA* H1N1 <1:10 <1:10 <1:10 <1:10 1:40 1:5,120

Sw/HK/8512/2001 EA H1N1 1:10,240 1:1,280 1:5,120 1:2,560 1:10,240 1:320

Sw/HK/72/2007 EA* H1N1 <1:10 <1:10 <1:10 <1:10 1:40 <1:10

Sw/HK/247/2009 EA* H1N1 <1:10 <1:10 <1:10 <1:10 1:40 1:2,560

Sw/HK/NS613/2009 EA* H1N1 <1:10 <1:10 <1:10 <1:10 1:10 1:2,560

Sw/HK/2481/2009 EA* H1N1 <1:10 <1:10 <1:10 <1:10 1:20 1:2,560

Sw/HK/NS186/2009 EA* H1N1 <1:10 <1:10 <1:10 <1:10 1:40 1:2,560

Ferret antisera

Sw/HK/1669/2002 EA H1N1 1:5,120 1:1,280 1:2,560 1:1,280 1:10,240 1:160

Sw/HK/NS129/2003 EA H1N1 1:5,120 1:1,280 1:2,560 1:1,280 1:10,240 1:160

Sw/HK/1716/2006 EA H1N1 1:2,560 1:640 1:1,280 1:640 1:2,560 <1:10

Sw/HK/NS952/2008 EA H1N1 1:2,560 1:640 1:640 1:320 1:2,560 <1:10

Figure 4 | Antigenic characterization of SwIV measured by haemagglutinininhibition assays. Titres are shaded according to their respective major SwIVHA lineages (see Figs 1–3); low (1:20, 1:40) and non-reactive titres (,1:10) areshaded in lighter colours. Underlined values represent homologous antibodytitres. EA-reassortant viruses (indicated by asterisks) showed poorcrossreactivity against antisera raised towards CS, TRIG, H1N1/2009 and late(2006–2009) ‘pure’ EA viruses. Excepting the earliest EA-reassortant virus (Sw/HK/72/2007), all remaining EA-reassortants reacted well against antiseraraised towards the EA-reassortant virus Sw/HK/1559/2009, indicatingprogressive antigenic change of this novel reassortant.

LETTER RESEARCH

2 6 M A Y 2 0 1 1 | V O L 4 7 3 | N A T U R E | 5 2 1

Macmillan Publishers Limited. All rights reserved©2011

Vijaykrishna et al Nature 2011

H1N1pdm emerged from out of

a surveillance gap

Page 6: Session 4: Modeling of potential pandemics

MERS

MERS emerged and remains in a surveillance gap.Rambaut, epidemic.bio.ed.ac.uk

Page 7: Session 4: Modeling of potential pandemics

Predicting emergence

and delayed virus shedding compared with pan-demic A/H1N1 virus.

Airborne transmission could be tested in asecond mammalian model system such as guineapigs (59), but this would still not provide con-clusive evidence that transmission among hu-mans would occur. The mutations we identifiedneed to be tested for their effect on transmissionin other A/H5N1 virus lineages (60), and exper-iments are needed to quantify how they affectviral fitness and virulence in birds and mammals.For pandemic preparedness, antiviral drugs andvaccine candidates against airborne-transmissiblevirus should be evaluated in depth. Mechanisticstudies on the phenotypic traits associated witheach of the identified amino acid substitutionsshould provide insights into the key determinantsof airborne virus transmission. Our findings in-dicate that HPAI A/H5N1 viruses have the po-tential to evolve directly to transmit by aerosol orrespiratory droplets between mammals, withoutreassortment in any intermediate host, and thuspose a risk of becoming pandemic in humans.Identification of the minimal requirements forvirus transmission between mammals may haveprognostic and diagnostic value for improvingpandemic preparedness (34).

References and Notes1. R. G. Webster, W. J. Bean, O. T. Gorman, T. M. Chambers,

Y. Kawaoka, Microbiol. Rev. 56, 152 (1992).2. P. Palese, M. L. Shaw, in Fields Virology, D. M. Knipe et al.,

Eds. (Lippincott Williams & Wilkins, Philadelphia, 2007),vol. 3, pp. 1647–1690.

3. P. F. Wright, G. Neumann, Y. Kawaoka, in Fields Virology,D. M. Knipe et al., Eds. (Lippincott Williams & Wilkins,Philadelphia, 2007), vol. 3, pp. 1691–1740.

4. W. Chen et al., Nat. Med. 7, 1306 (2001).5. G. M. Conenello, P. Palese, Cell Host Microbe 2, 207 (2007).6. R. A. Fouchier et al., J. Virol. 79, 2814 (2005).7. D. J. Alexander, Vet. Microbiol. 74, 3 (2000).8. D. J. Alexander, I. H. Brown, Rev. Sci. Tech. 28, 19 (2009).9. R. G. Webster, R. Rott, Cell 50, 665 (1987).

10. H. D. Klenk, W. Garten, Trends Microbiol. 2, 39 (1994).11. J. C. de Jong, E. C. Claas, A. D. Osterhaus, R. G. Webster,

W. L. Lim, Nature 389, 554 (1997).12. www.who.int/influenza/human_animal_interface/en/13. I. N. Kandun et al., N. Engl. J. Med. 355, 2186 (2006).14. K. Ungchusak et al., N. Engl. J. Med. 352, 333 (2005).15. H. Wang et al., Lancet 371, 1427 (2008).16. E. de Wit, Y. Kawaoka, M. D. de Jong, R. A. Fouchier,

Vaccine 26, D54 (2008).17. D. M. Tscherne, A. García-Sastre, J. Clin. Invest. 121, 6 (2011).18. S. Jackson et al., J. Virol. 83, 8131 (2009).19. T. R. Maines et al., Proc. Natl. Acad. Sci. U.S.A. 103,

12121 (2006).20. T. R. Maines et al., Virology 413, 139 (2011).21. E. M. Sorrell, H. Wan, Y. Araya, H. Song, D. R. Perez,

Proc. Natl. Acad. Sci. U.S.A. 106, 7565 (2009).22. E. M. Sorrell et al., Curr. Opin. Virol. 1, 635 (2011).23. H. L. Yen et al., J. Virol. 81, 6890 (2007).24. W. Smith, C. H. Andrewes, P. P. Laidlaw, Lancet 222, 66

(1933).25. See materials and methods and other supplementary

materials on Science Online.26. J. A. Maher, J. DeStefano, Lab Anim. 33, 50 (2004).27. V. J. Munster et al., Science 325, 481 (2009).28. T. Costa et al., Vet. Res. 43, 28 (2012).29. A. Mehle, J. A. Doudna, Proc. Natl. Acad. Sci. U.S.A. 106,

21312 (2009).30. J. Steel, A. C. Lowen, S. Mubareka, P. Palese, PLoS

Pathog. 5, e1000252 (2009).

31. N. Van Hoeven et al., Proc. Natl. Acad. Sci. U.S.A. 106,3366 (2009).

32. E. K. Subbarao, W. London, B. R. Murphy, J. Virol. 67,1761 (1993).

33. www.knaw.nl/Content/Internet_KNAW/publicaties/pdf/20071092.pdf

34. R. A. M. Fouchier, S. Herfst, A. D. M. E. Osterhaus,Science 335, 662 (2012).

35. S. Chutinimitkul et al., J. Virol. 84, 6825 (2010).36. R. J. Connor, Y. Kawaoka, R. G. Webster, J. C. Paulson,

Virology 205, 17 (1994).37. S. Yamada et al., Nature 444, 378 (2006).38. M. A. Nowak, Trends Ecol. Evol. 7, 118 (1992).39. R. Bodewes et al., Am. J. Pathol. 179, 30 (2011).40. E. J. Schrauwen et al., J. Virol. 86, 3975 (2012).41. S. L. Epstein, J. Infect. Dis. 193, 49 (2006).42. A. J. McMichael, F. M. Gotch, G. R. Noble, P. A. Beare,

N. Engl. J. Med. 309, 13 (1983).43. R. Bodewes et al., J. Virol. 85, 2695 (2011).44. J. H. Kreijtz et al., Vaccine 27, 4983 (2009).45. J. M. van den Brand et al., J. Infect. Dis. 201, 993 (2010).46. Y. Ha, D. J. Stevens, J. J. Skehel, D. C. Wiley, Proc. Natl.

Acad. Sci. U.S.A. 98, 11181 (2001).47. G. N. Rogers et al., J. Biol. Chem. 260, 7362 (1985).48. C. M. Deom, A. J. Caton, I. T. Schulze, Proc. Natl. Acad.

Sci. U.S.A. 83, 3771 (1986).49. S. Y. Mir-Shekari, D. A. Ashford, D. J. Harvey, R. A. Dwek,

I. T. Schulze, J. Biol. Chem. 272, 4027 (1997).50. I. A. Rudneva, N. A. Ilyushina, T. A. Timofeeva,

R. G. Webster, N. V. Kaverin, J. Gen. Virol. 86, 2831 (2005).51. J. Stevens et al., J. Mol. Biol. 381, 1382 (2008).52. M. Matrosovich et al., J. Virol. 74, 8502 (2000).53. Y. Bao et al., J. Virol. 82, 596 (2008).54. www.fao.org/avianflu/en/qanda.html55. A. Bataille, F. van der Meer, A. Stegeman, G. Koch,

PLoS Pathog. 7, e1002094 (2011).56. M. Jonges et al., J. Virol. 85, 10598 (2011).57. E. de Wit et al., J. Virol. 84, 1597 (2010).58. M. Imai et al., Nature 10.1038/nature10831 (2012).59. A. C. Lowen, S. Mubareka, T. M. Tumpey, A. García-Sastre,

P. Palese, Proc. Natl. Acad. Sci. U.S.A. 103, 9988 (2006).

60. WHO/OIE/FAO H5N1 Evolution Working Group, InfluenzaOther Respir. Viruses 3, 59 (2009).

Acknowledgments: We thank D. de Meulder, G. van Amerongen,and D. Akkermans for technical assistance. M. Peiris, Univ.of Hong Kong, provided A/Indonesia/5/2005 with permissionfrom I. Kandun of the Indonesian government. This work wasfinanced through NIAID-NIH contract HHSN266200700010C.D.J.S. and D.F.B. were supported in part by NIH Director’sPioneer Award DP1-OD000490-01. We acknowledge aNederlandse Organisatie voor Wetenschappelijk OnderzoekVICI grant, European Union FP7 program EMPERIE (223498),and Human Frontier Science Program grant P0050/2008.D.F.B. and D.J.S. acknowledge the use of the CamGriddistributed computing resource. Sequence data generatedfrom this study were deposited in GenBank with accessionnumbers CY116643 to CY116698. Special arrangements arein place with the NIH and the contractor at Mount Sinai Schoolof Medicine, New York, for sharing the viruses (and plasmids)in the present paper; please contact R.A.M.F. A.D.M. E.O.and G.F.R. are CSO and part-time employee of ViroClinicsBiosciences BV. A.D.M.E.O. has advisory affiliations on behalfof Viroclinics Biosciences BV with GlaxoSmithKline, Novartis,and Roche. A.D.M.E.O. and R.A.M.F. are holders of certificatesof shares in ViroClinics Biosciences B.V. To avoid any possibleconflict of interests, Erasmus MC policy dictates that the sharesas such are held by the Stichting Administratiekantoor ErasmusPersoneelsparticipaties. The board of this foundation is appointedby the Board of Governors of the Erasmus MC and exercises allvoting rights with regard to these shares.

Supplementary Materialswww.sciencemag.org/cgi/content/full/336/6088/1534/DC1Materials and MethodsSupplementary TextFigs. S1 to S10Tables S1 to S6References (61–72)

30 August 2011; accepted 31 May 201210.1126/science.1213362

REPORT

The Potential for RespiratoryDroplet–Transmissible A/H5N1 InfluenzaVirus to Evolve in a Mammalian HostColin A. Russell,1,2,3 Judith M. Fonville,1,2 AndréE. X. Brown,4 David F. Burke,1,2 David L. Smith,3,5,6

Sarah L. James,1,2 Sander Herfst,7 Sander van Boheemen,7 Martin Linster,7 Eefje J. Schrauwen,7

Leah Katzelnick,1,2 Ana Mosterín,1,2,8 Thijs Kuiken,7 Eileen Maher,9 Gabriele Neumann,9

Albert D. M. E. Osterhaus,7 Yoshihiro Kawaoka,9,10,11,12 Ron A. M. Fouchier,7 Derek J. Smith1,2,3,7*

Avian A/H5N1 influenza viruses pose a pandemic threat. As few as five amino acid substitutions,or four with reassortment, might be sufficient for mammal-to-mammal transmission through respiratorydroplets. From surveillance data, we found that two of these substitutions are common in A/H5N1viruses, and thus, some viruses might require only three additional substitutions to becometransmissible via respiratory droplets between mammals. We used a mathematical model of within-hostvirus evolution to study factors that could increase and decrease the probability of the remainingsubstitutions evolving after the virus has infected a mammalian host. These factors, combined with thepresence of some of these substitutions in circulating strains, make a virus evolving in nature apotentially serious threat. These results highlight critical areas in which more data are needed forassessing, and potentially averting, this threat.

Recent studies have shown that theA/Indonesia/5/2005 avian A/H5N1 influ-enza virus may require as few as five

amino acid substitutions (1), and the A/Vietnam/1203/2004 A/H5N1 influenza virus requires foursubstitutions and reassortment (2), to become

www.sciencemag.org SCIENCE VOL 336 22 JUNE 2012 1541

SPECIALSECTION

and delayed virus shedding compared with pan-demic A/H1N1 virus.

Airborne transmission could be tested in asecond mammalian model system such as guineapigs (59), but this would still not provide con-clusive evidence that transmission among hu-mans would occur. The mutations we identifiedneed to be tested for their effect on transmissionin other A/H5N1 virus lineages (60), and exper-iments are needed to quantify how they affectviral fitness and virulence in birds and mammals.For pandemic preparedness, antiviral drugs andvaccine candidates against airborne-transmissiblevirus should be evaluated in depth. Mechanisticstudies on the phenotypic traits associated witheach of the identified amino acid substitutionsshould provide insights into the key determinantsof airborne virus transmission. Our findings in-dicate that HPAI A/H5N1 viruses have the po-tential to evolve directly to transmit by aerosol orrespiratory droplets between mammals, withoutreassortment in any intermediate host, and thuspose a risk of becoming pandemic in humans.Identification of the minimal requirements forvirus transmission between mammals may haveprognostic and diagnostic value for improvingpandemic preparedness (34).

References and Notes1. R. G. Webster, W. J. Bean, O. T. Gorman, T. M. Chambers,

Y. Kawaoka, Microbiol. Rev. 56, 152 (1992).2. P. Palese, M. L. Shaw, in Fields Virology, D. M. Knipe et al.,

Eds. (Lippincott Williams & Wilkins, Philadelphia, 2007),vol. 3, pp. 1647–1690.

3. P. F. Wright, G. Neumann, Y. Kawaoka, in Fields Virology,D. M. Knipe et al., Eds. (Lippincott Williams & Wilkins,Philadelphia, 2007), vol. 3, pp. 1691–1740.

4. W. Chen et al., Nat. Med. 7, 1306 (2001).5. G. M. Conenello, P. Palese, Cell Host Microbe 2, 207 (2007).6. R. A. Fouchier et al., J. Virol. 79, 2814 (2005).7. D. J. Alexander, Vet. Microbiol. 74, 3 (2000).8. D. J. Alexander, I. H. Brown, Rev. Sci. Tech. 28, 19 (2009).9. R. G. Webster, R. Rott, Cell 50, 665 (1987).

10. H. D. Klenk, W. Garten, Trends Microbiol. 2, 39 (1994).11. J. C. de Jong, E. C. Claas, A. D. Osterhaus, R. G. Webster,

W. L. Lim, Nature 389, 554 (1997).12. www.who.int/influenza/human_animal_interface/en/13. I. N. Kandun et al., N. Engl. J. Med. 355, 2186 (2006).14. K. Ungchusak et al., N. Engl. J. Med. 352, 333 (2005).15. H. Wang et al., Lancet 371, 1427 (2008).16. E. de Wit, Y. Kawaoka, M. D. de Jong, R. A. Fouchier,

Vaccine 26, D54 (2008).17. D. M. Tscherne, A. García-Sastre, J. Clin. Invest. 121, 6 (2011).18. S. Jackson et al., J. Virol. 83, 8131 (2009).19. T. R. Maines et al., Proc. Natl. Acad. Sci. U.S.A. 103,

12121 (2006).20. T. R. Maines et al., Virology 413, 139 (2011).21. E. M. Sorrell, H. Wan, Y. Araya, H. Song, D. R. Perez,

Proc. Natl. Acad. Sci. U.S.A. 106, 7565 (2009).22. E. M. Sorrell et al., Curr. Opin. Virol. 1, 635 (2011).23. H. L. Yen et al., J. Virol. 81, 6890 (2007).24. W. Smith, C. H. Andrewes, P. P. Laidlaw, Lancet 222, 66

(1933).25. See materials and methods and other supplementary

materials on Science Online.26. J. A. Maher, J. DeStefano, Lab Anim. 33, 50 (2004).27. V. J. Munster et al., Science 325, 481 (2009).28. T. Costa et al., Vet. Res. 43, 28 (2012).29. A. Mehle, J. A. Doudna, Proc. Natl. Acad. Sci. U.S.A. 106,

21312 (2009).30. J. Steel, A. C. Lowen, S. Mubareka, P. Palese, PLoS

Pathog. 5, e1000252 (2009).

31. N. Van Hoeven et al., Proc. Natl. Acad. Sci. U.S.A. 106,3366 (2009).

32. E. K. Subbarao, W. London, B. R. Murphy, J. Virol. 67,1761 (1993).

33. www.knaw.nl/Content/Internet_KNAW/publicaties/pdf/20071092.pdf

34. R. A. M. Fouchier, S. Herfst, A. D. M. E. Osterhaus,Science 335, 662 (2012).

35. S. Chutinimitkul et al., J. Virol. 84, 6825 (2010).36. R. J. Connor, Y. Kawaoka, R. G. Webster, J. C. Paulson,

Virology 205, 17 (1994).37. S. Yamada et al., Nature 444, 378 (2006).38. M. A. Nowak, Trends Ecol. Evol. 7, 118 (1992).39. R. Bodewes et al., Am. J. Pathol. 179, 30 (2011).40. E. J. Schrauwen et al., J. Virol. 86, 3975 (2012).41. S. L. Epstein, J. Infect. Dis. 193, 49 (2006).42. A. J. McMichael, F. M. Gotch, G. R. Noble, P. A. Beare,

N. Engl. J. Med. 309, 13 (1983).43. R. Bodewes et al., J. Virol. 85, 2695 (2011).44. J. H. Kreijtz et al., Vaccine 27, 4983 (2009).45. J. M. van den Brand et al., J. Infect. Dis. 201, 993 (2010).46. Y. Ha, D. J. Stevens, J. J. Skehel, D. C. Wiley, Proc. Natl.

Acad. Sci. U.S.A. 98, 11181 (2001).47. G. N. Rogers et al., J. Biol. Chem. 260, 7362 (1985).48. C. M. Deom, A. J. Caton, I. T. Schulze, Proc. Natl. Acad.

Sci. U.S.A. 83, 3771 (1986).49. S. Y. Mir-Shekari, D. A. Ashford, D. J. Harvey, R. A. Dwek,

I. T. Schulze, J. Biol. Chem. 272, 4027 (1997).50. I. A. Rudneva, N. A. Ilyushina, T. A. Timofeeva,

R. G. Webster, N. V. Kaverin, J. Gen. Virol. 86, 2831 (2005).51. J. Stevens et al., J. Mol. Biol. 381, 1382 (2008).52. M. Matrosovich et al., J. Virol. 74, 8502 (2000).53. Y. Bao et al., J. Virol. 82, 596 (2008).54. www.fao.org/avianflu/en/qanda.html55. A. Bataille, F. van der Meer, A. Stegeman, G. Koch,

PLoS Pathog. 7, e1002094 (2011).56. M. Jonges et al., J. Virol. 85, 10598 (2011).57. E. de Wit et al., J. Virol. 84, 1597 (2010).58. M. Imai et al., Nature 10.1038/nature10831 (2012).59. A. C. Lowen, S. Mubareka, T. M. Tumpey, A. García-Sastre,

P. Palese, Proc. Natl. Acad. Sci. U.S.A. 103, 9988 (2006).

60. WHO/OIE/FAO H5N1 Evolution Working Group, InfluenzaOther Respir. Viruses 3, 59 (2009).

Acknowledgments: We thank D. de Meulder, G. van Amerongen,and D. Akkermans for technical assistance. M. Peiris, Univ.of Hong Kong, provided A/Indonesia/5/2005 with permissionfrom I. Kandun of the Indonesian government. This work wasfinanced through NIAID-NIH contract HHSN266200700010C.D.J.S. and D.F.B. were supported in part by NIH Director’sPioneer Award DP1-OD000490-01. We acknowledge aNederlandse Organisatie voor Wetenschappelijk OnderzoekVICI grant, European Union FP7 program EMPERIE (223498),and Human Frontier Science Program grant P0050/2008.D.F.B. and D.J.S. acknowledge the use of the CamGriddistributed computing resource. Sequence data generatedfrom this study were deposited in GenBank with accessionnumbers CY116643 to CY116698. Special arrangements arein place with the NIH and the contractor at Mount Sinai Schoolof Medicine, New York, for sharing the viruses (and plasmids)in the present paper; please contact R.A.M.F. A.D.M. E.O.and G.F.R. are CSO and part-time employee of ViroClinicsBiosciences BV. A.D.M.E.O. has advisory affiliations on behalfof Viroclinics Biosciences BV with GlaxoSmithKline, Novartis,and Roche. A.D.M.E.O. and R.A.M.F. are holders of certificatesof shares in ViroClinics Biosciences B.V. To avoid any possibleconflict of interests, Erasmus MC policy dictates that the sharesas such are held by the Stichting Administratiekantoor ErasmusPersoneelsparticipaties. The board of this foundation is appointedby the Board of Governors of the Erasmus MC and exercises allvoting rights with regard to these shares.

Supplementary Materialswww.sciencemag.org/cgi/content/full/336/6088/1534/DC1Materials and MethodsSupplementary TextFigs. S1 to S10Tables S1 to S6References (61–72)

30 August 2011; accepted 31 May 201210.1126/science.1213362

REPORT

The Potential for RespiratoryDroplet–Transmissible A/H5N1 InfluenzaVirus to Evolve in a Mammalian HostColin A. Russell,1,2,3 Judith M. Fonville,1,2 AndréE. X. Brown,4 David F. Burke,1,2 David L. Smith,3,5,6

Sarah L. James,1,2 Sander Herfst,7 Sander van Boheemen,7 Martin Linster,7 Eefje J. Schrauwen,7

Leah Katzelnick,1,2 Ana Mosterín,1,2,8 Thijs Kuiken,7 Eileen Maher,9 Gabriele Neumann,9

Albert D. M. E. Osterhaus,7 Yoshihiro Kawaoka,9,10,11,12 Ron A. M. Fouchier,7 Derek J. Smith1,2,3,7*

Avian A/H5N1 influenza viruses pose a pandemic threat. As few as five amino acid substitutions,or four with reassortment, might be sufficient for mammal-to-mammal transmission through respiratorydroplets. From surveillance data, we found that two of these substitutions are common in A/H5N1viruses, and thus, some viruses might require only three additional substitutions to becometransmissible via respiratory droplets between mammals. We used a mathematical model of within-hostvirus evolution to study factors that could increase and decrease the probability of the remainingsubstitutions evolving after the virus has infected a mammalian host. These factors, combined with thepresence of some of these substitutions in circulating strains, make a virus evolving in nature apotentially serious threat. These results highlight critical areas in which more data are needed forassessing, and potentially averting, this threat.

Recent studies have shown that theA/Indonesia/5/2005 avian A/H5N1 influ-enza virus may require as few as five

amino acid substitutions (1), and the A/Vietnam/1203/2004 A/H5N1 influenza virus requires foursubstitutions and reassortment (2), to become

www.sciencemag.org SCIENCE VOL 336 22 JUNE 2012 1541

SPECIALSECTION

(10−5 mutations per site, per genome replication),the initial virus population expands exponentially[each infected cell produces 104 virions (11, 12),and 1010 cells can be infected (13)] until it reaches1014 virions, after which the virus population sizestays roughly constant, and selection is modeledby use of differences in expected numbers ofprogeny (fig. S2 and table S2) (6). The results

of the model are largely insensitive to thenumber of cells that can be infected, maximumvirus population size, and whether the viruspopulation remains roughly constant or declines(figs. S3 to S5). Typical infections were simu-lated out to 5 days corresponding to theapproximate time of peak viral load, and long-duration infections to 14 days (14).

It is not possible to calculate the level of riskprecisely because of uncertainties in some as-pects of the biology. We used the model to com-pare the relative effects of factors that couldincrease or decrease the probability of accumu-lating mutations and to identify areas for furtherinvestigation that are critical for more accuraterisk assessment. We compare and contrast the ef-fects of factors that can increase the probability ofaccumulating mutations and thus evolving a res-piratory droplet–transmissible A/H5N1 influenzavirus in a mammalian host, and factors that coulddecrease the probability of evolving a such a vi-rus. The factors we considered that can increasethe probability are random mutation, positive se-lection, long infection, alternate functionally equiv-alent substitutions, and transmission of partiallyadapted viruses as a proportion of the within-host diversity both in the avian-mammal and themammal-mammal transmission events (10, 14–18).The factors we considered that can decreasethe probability are an effective immune response,deleterious substitutions, and order-dependencein the acquisition of substitutions. We consideredthese factors for starting viruses differing inthe number of mutations that separates themfrom a respiratory droplet–transmissible A/H5N1virus—viruses that require five, four, three, two,or one mutations at specific positions in the vi-rus HA, reflecting that zero, one, two, three, orfour of the mutations are already present in theavian population and thus are present at the startof the infection inmammals.We treat each aminoacid substitution as if it can be acquired by asingle-nucleotide mutation, as is the case for

0 1 2 3 4 5

hill-climbhill-climb + deleterious

hill-climb + order

0 1 2 3 4 5

hill-climbhill-climb + intra-host diversity

hill-climb + functionally equiv. mutations

0 1 2 3 4 5−25

−20

−15

−10

−5

0

hill-climball-or-nothing

neutral

Time (days)

log 1

0(pr

opor

tion

of m

utan

t viru

s in

tota

l viru

s po

pula

tion)

A B

Time (days) Time (days)

C

Fig. 3. Factors that increase or decrease the proportion of respiratorydroplet–transmissible A/H5N1 virus based on starting viruses that require five(blue), four (green), three (orange), two (red), or one (purple) mutation (ormutations) to become respiratory droplet–transmissible. (A) The effect of hill-climb and all-or-nothing positive selection compared with random mutationalone. (B) The effect of avian–mammal transmission of partially adapted virusas a result of intra-host diversity (100 viruses start the infection, one ofwhich has a mutation) and the effect of alternate substitutions with 10functionally equivalent sites for a virus requiring five mutations (blue), nine

sites for a virus requiring four mutations (green), eight sites for a virusrequiring three mutations (orange), seven sites for a virus requiring twomutations (red), and six sites for a virus requiring one mutation (purple),both with hill-climb selection, compared with hill-climb selection alone. (C)The effect of two of the required substitutions being individually deleterious(for these two specific substitutions, either substitution alone reduces thereplicative fitness of the virus to zero) and the effect of complete orderdependence of acquiring substitutions, both with hill-climb selection ascompared with hill-climb selection alone.

0 1 2 3 4 5−10

−5

0

5

10

15

0 1 2 3 4 5−25

−20

−15

−10

−5

0

Time (days)

log 1

0(pr

opor

tion

of m

uta n

t viru

s in

tota

l viru

s po

pula

tion)

A

lo

g 10(

expe

cted

num

ber

of m

utan

t viru

ses)

B

Time (days)

Fig. 2. Expected absolute numbers and proportions of respiratory droplet–transmissible A/H5N1 virionswithin a host initially infected by strains that require five (blue), four (green), three (orange), two (red), orone (purple) mutation (or mutations) to become respiratory droplet–transmissible, calculated from thedeterministic model. (A) The absolute number of respiratory droplet–transmissible A/H5N1 viruses in ahost. The intersections with the gray line indicate the point when at least one virus in each host is expectedto have the required mutations. The change in slope is due to the transition in the virus population fromexponential expansion to constant size. (B) Expected proportion of respiratory droplet–transmissibleA/H5N1 viruses in the total virus population over time in the random mutation case (when all mutationsare fitness-neutral).

22 JUNE 2012 VOL 336 SCIENCE www.sciencemag.org1544

H5N1

“It is not possible to calculate the level of risk precisely because of uncertainties in some aspects of the biology.”

Aim: basic science, not risk assessment.

cf.: experience with HIV, Bonhoeffer et al Science 2014, Kouyos et al PLOS Genetics 2012, Carlson et al Science 2014, Fraser et al Science 2014

Page 8: Session 4: Modeling of potential pandemics

whose fates have been followed since 1971. Tissue sampling for DNA analyses started in1988. Blood samples were taken from all captured sheep until 1993 and stored inpreservative at220 8C. Sampling resumed in 1997, when hair samples were taken from allcaptured sheep by plucking 50–100 hairs including roots from the back or flank. Hairswere kept either in paper envelopes or plastic bags containing about 5 g of silica at roomtemperature. From 1998 to 2002, a tissue sample from each captured sheepwas taken fromthe ear with an 8-mm punch. Ear tissue was kept at 220 8C in a solution of 20%dimethylsulphoxide saturated with NaCl. We sampled 433 marked individuals over thecourse of the study.

DNAwas extracted from blood with a standard phenol–chloroformmethod, and fromeither 20–30 hairs including follicles or about 5mg of ear tissue, using the QIAamp tissueextraction kit (Qiagen Inc., Mississauga, Ontario). Polymerase chain reactionamplification at 20 ungulate-derived microsatellite loci, 15 as described previously5 plusMCM527, BM4025, MAF64, OarFCB193 and MAF92 (refs 25, 26), and fragment analysiswere performed as described elsewhere5. After correction for multiple comparisons, wefound no evidence for allelic or genotypic disequilibria at or among these 20 loci.

Paternity of 241 individuals was assigned by using the likelihood-based approachdescribed in CERVUS27 at a confidence level of more than 95% with input parametersgiven in ref. 5. After paternity analysis, we used KINSHIP28 to identify 31 clusters of 104paternal half-sibs among the unassigned offspring. A paternal half-sibship consisted of allpairs of individuals of unassigned paternity that were identified in the KINSHIP analysis ashaving a likelihood ratio of the probability of a paternal half-sib relationship versusunrelated with an associated P , 0.05 (ref. 28). Members of reconstructed paternal half-sibships were assigned a common unknown paternal identity for the animal modelanalyses. Paternal identity links in the pedigree were therefore defined for 345 individuals.

Animal model analysesBreeding values, genetic variance components and heritabilities were estimated by using amultiple trait restricted-estimate maximum-likelihood (REML) model implemented bythe programs PEST29 and VCE30. An animal model was fitted in which the phenotype ofeach animal was broken down into components of additive genetic value and otherrandom and fixed effects: y ¼ Xb þ Za þ Pc þ e, where y was a vector of phenotypicvalues, b was a vector of fixed effects, a and c were vectors of additive genetic andpermanent environmental, e was a vector of residual values, and X, Z and P were thecorresponding design matrices relating records to the appropriate fixed or randomeffects18. Fixed effects included age (factor) and the average weight of yearling ewes in theyear of measurement (covariate), which is a better index of resource availability thanpopulation size because it accounts for time-lagged effects4. The permanentenvironmental effect grouped repeated observations on the same individual to quantifyany remaining between-individual variance over and above that due to additive geneticeffects, which would be due to maternal or other long-term environmental and non-additive genetic effects.

The total phenotypic variance (Vp) was therefore partitioned into three components:the additive genetic variance (Va), the permanent environmental variance (Ve) and theresidual variance (Vr), thus: Vp ¼ Va þ Ve þ Vr. Heritability was calculated as h2 ¼ Va/Vp. The VCE

30 program returns standard errors on all variance components and ratios.Best linear unbiased predictors of individual breeding values were quantified by usingREML estimates of the variance components obtained with PEST29. All statistical testswere conducted in SPLUS 6.1.

Received 11 August; accepted 17 October 2003; doi:10.1038/nature02177.

1. Thelen, T. H. Effects of harvest on antlers of simulated populations of elk. J. Wild. Mgmt 55, 243–249

(1991).

2. Harris, R. B., Wall, W. A. & Allendorf, F. W. Genetic consequences of hunting: what do we know and

what should we do? Wildl. Soc. Bull. 30, 634–643 (2002).

3. Festa-Bianchet, M. in Animal Behavior and Wildlife Conservation (eds Apollonio, M. &

Festa-Bianchet, M.) 191–207 (Island Press, Washington DC, 2003).

4. Festa-Bianchet, M., Coltman, D. W., Turelli, L. & Jorgenson, J. T. Relative allocation to horn and body

growth in bighorn rams varies with resource availability. Behav. Ecol. (in the press).

5. Coltman, D.W., Festa-Bianchet, M., Jorgenson, J. T. & Strobeck, C. Age-dependent sexual selection in

bighorn rams. Proc. R. Soc. Lond. B 269, 165–172 (2002).

6. Milner-Gulland, E. J. et al.Conservation—reproductive collapse in saiga antelope harems.Nature 422,

135 (2003).

7. Marty, S. Sacrificial ram. Can. Geographic November/December, 37–50 (2002).

8. Clutton-Brock, T. H., Coulson, T. N., Milner-Gulland, E. J., Thomson, D. & Armstrong, H. M. Sex

differences in emigration and mortality affect optimal management of deer populations. Nature 415,

633–637 (2002).

9. Ginsberg, J. R. &Milner Gulland, E. J. Sex-biased harvesting and population dynamics in ungulates—

implications for conservation and sustainable use. Conserv. Biol. 8, 157–166 (1994).

10. Langvatn, R. & Loison, A. Consequences of harvesting on age structure, sex ratio and population

dynamics of red deer Cervus elaphus in central Norway. Wildl. Biol. 5, 213–223 (1999).

11. Laurian, C., Ouellet, J. P., Courtois, R., Breton, L. & St-Onge, S. Effects of intensive harvesting on

moose reproduction. J. Appl. Ecol. 37, 515–531 (2000).

12. Ashley, M. V. et al. Evolutionarily enlightened management. Biol. Conserv. 111, 115–123 (2003).

13. Conover, D. O. & Munch, S. B. Sustaining fisheries yields over evolutionary time scales. Science 297,

94–96 (2002).

14. Palumbi, S. R. Evolution—humans as the world’s greatest evolutionary force. Science 293, 1786–1790

(2001).

15. Law, R. Fishing, selection, and phenotypic evolution. Ices J. Mar. Sci. 57, 659–668 (2000).

16. Jachmann, H., Berry, P. S.M. & Imae, H. Tusklessness in African elephants—a future trend.Afr. J. Ecol.

33, 230–235 (1995).

17. Jorgenson, J. T., Festa-Bianchet, M. & Wishart, W. D. Effects of population density on horn

development in bighorn rams. J. Wildl. Mgmt 62, 1011–1020 (1998).

18. Lynch, M. & Walsh, B. Genetics and Analysis of Quantitative Traits (Massachusetts, Sinauer,

Sunderland, 1998).

19. Hogg, J. T. & Forbes, S. H. Mating in bighorn sheep: frequent male reproduction via a high-risk

‘unconventional’ tactic. Behav. Ecol. Sociobiol. 41, 33–48 (1997).

20. Hendry, A. P. & Kinnison, M. T. Perspective: The pace of modern life: measuring rates of

contemporary microevolution. Evolution 53, 1637–1653 (1999).

21. Merila, J., Kruuk, L. E. B. & Sheldon, B. C. Cryptic evolution in a wild bird population. Nature 412,

76–79 (2001).

22. Coltman, D. W., Pilkington, J., Kruuk, L. E. B., Wilson, K. & Pemberton, J. M. Positive genetic

correlation between parasite resistance and body size in a free-living ungulate population. Evolution

55, 2116–2125 (2001).

23. Jorgenson, J. T., Festa-Bianchet, M. & Wishart, W. D. Harvesting bighorn ewes—consequences for

population-size and trophy ram production. J. Wildl. Mgmt 57, 429–435 (1993).

24. Festa-Bianchet, M., Jorgenson, J. T., King, W. J., Smith, K. G. & Wishart, W. D. The development of

sexual dimorphism: seasonal and lifetime mass changes in bighorn sheep. Can. J. Zool. 74, 330–342

(1996).

25. Crawford, A. M. et al. An autosomal genetic linkage map of the sheep genome.Genetics 140, 703–724

(1995).

26. Slate, J. et al. Bovine microsatellite loci are highly conserved in red deer (Cervus elaphus), sika deer

(Cervus nippon) and Soay sheep (Ovis aries). Anim. Genet. 29, 307–315 (1998).

27. Marshall, T. C., Slate, J., Kruuk, L. E. B. & Pemberton, J. M. Statistical confidence for likelihood-based

paternity inference in natural populations. Mol. Ecol. 7, 639–655 (1998).

28. Goodnight, K. F. & Queller, D. C. Computer software for performing likelihood tests of pedigree

relationship using genetic markers. Mol. Ecol. 8, 1231–1234 (1999).

29. Groeneveld, E., Kovac, M., Wang, T. L. & Fernando, R. L. Computing algorithms in a general purpose

BLUP package for multivariate prediction and estimation. Arch. Anim. Breed. 15, 399–412 (1992).

30. Groeneveld, E. User’s Guide: REMLVCE, a Multivariate Multi-Model Restricted Maximum Likelihood

(Co)variance Estimation Package, Version 3.2 (Institute of Animal Husbandry and Animal Behaviour,

Federal Research Center of Agriculture (FAL) (Mariensee, Germany, 1995).

Acknowledgements We thank the many students, colleagues, volunteers and assistants thatcontributed to this research over the past 30 years. B.Wishart initiated the RamMountain project.Our research was funded by the Alberta Conservation Association, Alberta Fish and WildlifeDivision, Alberta Recreation, Sports, Parks and Wildlife Foundation, Eppley Foundation forResearch, Foundation for North American Wild Sheep, National Geographic Society, NaturalEnvironment Research Council (UK), Natural Sciences and Engineering Research Council ofCanada, Rocky Mountain Elk Foundation (Canada), and the Universite de Sherbrooke. We aregrateful for the logistical support of the Alberta Forest Service.

Competing interests statement The authors declare that they have no competing financialinterests.

Correspondence and requests for materials should be addressed to D.W.C.([email protected]).

..............................................................

The role of evolution in theemergence of infectious diseasesRustom Antia1, Roland R. Regoes1, Jacob C. Koella2 & Carl T. Bergstrom3

1Department of Biology, Emory University, Atlanta, Georgia 30322, USA2Laboratoire de Parasitologie Evolutive, Universite Pierre et Marie Curie,75252 Paris, France3Department of Biology, University of Washington, Seattle, Washington 98195,USA.............................................................................................................................................................................

It is unclear when, where and how novel pathogens such ashuman immunodeficiency virus (HIV), monkeypox and severeacute respiratory syndrome (SARS) will cross the barriers thatseparate their natural reservoirs from human populations andignite the epidemic spread of novel infectious diseases. Newpathogens are believed to emerge from animal reservoirs whenecological changes increase the pathogen’s opportunities to enterthe human population1 and to generate subsequent human-to-human transmission2. Effective human-to-human transmissionrequires that the pathogen’s basic reproductive number, R0,should exceed one, where R0 is the average number of secondaryinfections arising from one infected individual in a completelysusceptible population3. However, an increase in R0, even wheninsufficient to generate an epidemic, nonetheless increases thenumber of subsequently infected individuals. Here we show that,

letters to nature

NATURE | VOL 426 | 11 DECEMBER 2003 | www.nature.com/nature658 © 2003 Nature Publishing Group

whose fates have been followed since 1971. Tissue sampling for DNA analyses started in1988. Blood samples were taken from all captured sheep until 1993 and stored inpreservative at220 8C. Sampling resumed in 1997, when hair samples were taken from allcaptured sheep by plucking 50–100 hairs including roots from the back or flank. Hairswere kept either in paper envelopes or plastic bags containing about 5 g of silica at roomtemperature. From 1998 to 2002, a tissue sample from each captured sheepwas taken fromthe ear with an 8-mm punch. Ear tissue was kept at 220 8C in a solution of 20%dimethylsulphoxide saturated with NaCl. We sampled 433 marked individuals over thecourse of the study.

DNAwas extracted from blood with a standard phenol–chloroformmethod, and fromeither 20–30 hairs including follicles or about 5mg of ear tissue, using the QIAamp tissueextraction kit (Qiagen Inc., Mississauga, Ontario). Polymerase chain reactionamplification at 20 ungulate-derived microsatellite loci, 15 as described previously5 plusMCM527, BM4025, MAF64, OarFCB193 and MAF92 (refs 25, 26), and fragment analysiswere performed as described elsewhere5. After correction for multiple comparisons, wefound no evidence for allelic or genotypic disequilibria at or among these 20 loci.

Paternity of 241 individuals was assigned by using the likelihood-based approachdescribed in CERVUS27 at a confidence level of more than 95% with input parametersgiven in ref. 5. After paternity analysis, we used KINSHIP28 to identify 31 clusters of 104paternal half-sibs among the unassigned offspring. A paternal half-sibship consisted of allpairs of individuals of unassigned paternity that were identified in the KINSHIP analysis ashaving a likelihood ratio of the probability of a paternal half-sib relationship versusunrelated with an associated P , 0.05 (ref. 28). Members of reconstructed paternal half-sibships were assigned a common unknown paternal identity for the animal modelanalyses. Paternal identity links in the pedigree were therefore defined for 345 individuals.

Animal model analysesBreeding values, genetic variance components and heritabilities were estimated by using amultiple trait restricted-estimate maximum-likelihood (REML) model implemented bythe programs PEST29 and VCE30. An animal model was fitted in which the phenotype ofeach animal was broken down into components of additive genetic value and otherrandom and fixed effects: y ¼ Xb þ Za þ Pc þ e, where y was a vector of phenotypicvalues, b was a vector of fixed effects, a and c were vectors of additive genetic andpermanent environmental, e was a vector of residual values, and X, Z and P were thecorresponding design matrices relating records to the appropriate fixed or randomeffects18. Fixed effects included age (factor) and the average weight of yearling ewes in theyear of measurement (covariate), which is a better index of resource availability thanpopulation size because it accounts for time-lagged effects4. The permanentenvironmental effect grouped repeated observations on the same individual to quantifyany remaining between-individual variance over and above that due to additive geneticeffects, which would be due to maternal or other long-term environmental and non-additive genetic effects.

The total phenotypic variance (Vp) was therefore partitioned into three components:the additive genetic variance (Va), the permanent environmental variance (Ve) and theresidual variance (Vr), thus: Vp ¼ Va þ Ve þ Vr. Heritability was calculated as h2 ¼ Va/Vp. The VCE

30 program returns standard errors on all variance components and ratios.Best linear unbiased predictors of individual breeding values were quantified by usingREML estimates of the variance components obtained with PEST29. All statistical testswere conducted in SPLUS 6.1.

Received 11 August; accepted 17 October 2003; doi:10.1038/nature02177.

1. Thelen, T. H. Effects of harvest on antlers of simulated populations of elk. J. Wild. Mgmt 55, 243–249

(1991).

2. Harris, R. B., Wall, W. A. & Allendorf, F. W. Genetic consequences of hunting: what do we know and

what should we do? Wildl. Soc. Bull. 30, 634–643 (2002).

3. Festa-Bianchet, M. in Animal Behavior and Wildlife Conservation (eds Apollonio, M. &

Festa-Bianchet, M.) 191–207 (Island Press, Washington DC, 2003).

4. Festa-Bianchet, M., Coltman, D. W., Turelli, L. & Jorgenson, J. T. Relative allocation to horn and body

growth in bighorn rams varies with resource availability. Behav. Ecol. (in the press).

5. Coltman, D.W., Festa-Bianchet, M., Jorgenson, J. T. & Strobeck, C. Age-dependent sexual selection in

bighorn rams. Proc. R. Soc. Lond. B 269, 165–172 (2002).

6. Milner-Gulland, E. J. et al.Conservation—reproductive collapse in saiga antelope harems.Nature 422,

135 (2003).

7. Marty, S. Sacrificial ram. Can. Geographic November/December, 37–50 (2002).

8. Clutton-Brock, T. H., Coulson, T. N., Milner-Gulland, E. J., Thomson, D. & Armstrong, H. M. Sex

differences in emigration and mortality affect optimal management of deer populations. Nature 415,

633–637 (2002).

9. Ginsberg, J. R. &Milner Gulland, E. J. Sex-biased harvesting and population dynamics in ungulates—

implications for conservation and sustainable use. Conserv. Biol. 8, 157–166 (1994).

10. Langvatn, R. & Loison, A. Consequences of harvesting on age structure, sex ratio and population

dynamics of red deer Cervus elaphus in central Norway. Wildl. Biol. 5, 213–223 (1999).

11. Laurian, C., Ouellet, J. P., Courtois, R., Breton, L. & St-Onge, S. Effects of intensive harvesting on

moose reproduction. J. Appl. Ecol. 37, 515–531 (2000).

12. Ashley, M. V. et al. Evolutionarily enlightened management. Biol. Conserv. 111, 115–123 (2003).

13. Conover, D. O. & Munch, S. B. Sustaining fisheries yields over evolutionary time scales. Science 297,

94–96 (2002).

14. Palumbi, S. R. Evolution—humans as the world’s greatest evolutionary force. Science 293, 1786–1790

(2001).

15. Law, R. Fishing, selection, and phenotypic evolution. Ices J. Mar. Sci. 57, 659–668 (2000).

16. Jachmann, H., Berry, P. S.M. & Imae, H. Tusklessness in African elephants—a future trend.Afr. J. Ecol.

33, 230–235 (1995).

17. Jorgenson, J. T., Festa-Bianchet, M. & Wishart, W. D. Effects of population density on horn

development in bighorn rams. J. Wildl. Mgmt 62, 1011–1020 (1998).

18. Lynch, M. & Walsh, B. Genetics and Analysis of Quantitative Traits (Massachusetts, Sinauer,

Sunderland, 1998).

19. Hogg, J. T. & Forbes, S. H. Mating in bighorn sheep: frequent male reproduction via a high-risk

‘unconventional’ tactic. Behav. Ecol. Sociobiol. 41, 33–48 (1997).

20. Hendry, A. P. & Kinnison, M. T. Perspective: The pace of modern life: measuring rates of

contemporary microevolution. Evolution 53, 1637–1653 (1999).

21. Merila, J., Kruuk, L. E. B. & Sheldon, B. C. Cryptic evolution in a wild bird population. Nature 412,

76–79 (2001).

22. Coltman, D. W., Pilkington, J., Kruuk, L. E. B., Wilson, K. & Pemberton, J. M. Positive genetic

correlation between parasite resistance and body size in a free-living ungulate population. Evolution

55, 2116–2125 (2001).

23. Jorgenson, J. T., Festa-Bianchet, M. & Wishart, W. D. Harvesting bighorn ewes—consequences for

population-size and trophy ram production. J. Wildl. Mgmt 57, 429–435 (1993).

24. Festa-Bianchet, M., Jorgenson, J. T., King, W. J., Smith, K. G. & Wishart, W. D. The development of

sexual dimorphism: seasonal and lifetime mass changes in bighorn sheep. Can. J. Zool. 74, 330–342

(1996).

25. Crawford, A. M. et al. An autosomal genetic linkage map of the sheep genome.Genetics 140, 703–724

(1995).

26. Slate, J. et al. Bovine microsatellite loci are highly conserved in red deer (Cervus elaphus), sika deer

(Cervus nippon) and Soay sheep (Ovis aries). Anim. Genet. 29, 307–315 (1998).

27. Marshall, T. C., Slate, J., Kruuk, L. E. B. & Pemberton, J. M. Statistical confidence for likelihood-based

paternity inference in natural populations. Mol. Ecol. 7, 639–655 (1998).

28. Goodnight, K. F. & Queller, D. C. Computer software for performing likelihood tests of pedigree

relationship using genetic markers. Mol. Ecol. 8, 1231–1234 (1999).

29. Groeneveld, E., Kovac, M., Wang, T. L. & Fernando, R. L. Computing algorithms in a general purpose

BLUP package for multivariate prediction and estimation. Arch. Anim. Breed. 15, 399–412 (1992).

30. Groeneveld, E. User’s Guide: REMLVCE, a Multivariate Multi-Model Restricted Maximum Likelihood

(Co)variance Estimation Package, Version 3.2 (Institute of Animal Husbandry and Animal Behaviour,

Federal Research Center of Agriculture (FAL) (Mariensee, Germany, 1995).

Acknowledgements We thank the many students, colleagues, volunteers and assistants thatcontributed to this research over the past 30 years. B.Wishart initiated the RamMountain project.Our research was funded by the Alberta Conservation Association, Alberta Fish and WildlifeDivision, Alberta Recreation, Sports, Parks and Wildlife Foundation, Eppley Foundation forResearch, Foundation for North American Wild Sheep, National Geographic Society, NaturalEnvironment Research Council (UK), Natural Sciences and Engineering Research Council ofCanada, Rocky Mountain Elk Foundation (Canada), and the Universite de Sherbrooke. We aregrateful for the logistical support of the Alberta Forest Service.

Competing interests statement The authors declare that they have no competing financialinterests.

Correspondence and requests for materials should be addressed to D.W.C.([email protected]).

..............................................................

The role of evolution in theemergence of infectious diseasesRustom Antia1, Roland R. Regoes1, Jacob C. Koella2 & Carl T. Bergstrom3

1Department of Biology, Emory University, Atlanta, Georgia 30322, USA2Laboratoire de Parasitologie Evolutive, Universite Pierre et Marie Curie,75252 Paris, France3Department of Biology, University of Washington, Seattle, Washington 98195,USA.............................................................................................................................................................................

It is unclear when, where and how novel pathogens such ashuman immunodeficiency virus (HIV), monkeypox and severeacute respiratory syndrome (SARS) will cross the barriers thatseparate their natural reservoirs from human populations andignite the epidemic spread of novel infectious diseases. Newpathogens are believed to emerge from animal reservoirs whenecological changes increase the pathogen’s opportunities to enterthe human population1 and to generate subsequent human-to-human transmission2. Effective human-to-human transmissionrequires that the pathogen’s basic reproductive number, R0,should exceed one, where R0 is the average number of secondaryinfections arising from one infected individual in a completelysusceptible population3. However, an increase in R0, even wheninsufficient to generate an epidemic, nonetheless increases thenumber of subsequently infected individuals. Here we show that,

letters to nature

NATURE | VOL 426 | 11 DECEMBER 2003 | www.nature.com/nature658 © 2003 Nature Publishing Group

as a consequence of this, the probability of pathogen evolution toR0 > 1 and subsequent disease emergence can increase markedly.

The emergence of a disease combines two elements: the intro-duction of the pathogen into the human population and itssubsequent spread and maintenance within the population. Eco-logical factors such as human behaviour can influence both of theseelements, and consequently ecology has been recognized to have animportant role in the emergence of disease1,2,4. In contrast, evol-utionary factors including the adaptation of the pathogen to growthwithin humans and the subsequent transmission of the pathogenbetween humans are mostly considered in terms of changes in thevirulence of the pathogen, and are often thought to have a lesser rolein the initial emergence of pathogens4. One exception5 suggests thatimmunocompromised individuals might provide “stepping stones”for the evolution of pathogens.

The successful emergence of a pathogen requiresR0 to exceed onein the new host. Only then can an introduction trigger emergence2.(Here we use R0 to refer to spread in human populations, not in thenatural reservoir.) If R0 for a potential pathogen exceeds one, thisscenario represents an epidemic waiting to happen. By contrast,when R0 is initially less than one, infections will inevitably die outand there will be no epidemic unless genetic or ecological changesdrive R0 above one.

There are a number of ways in which R0 can increase. Ecologicalchanges such as changes in host density or behaviour can increaseR0, as can genetic changes in the pathogen population or in thepopulation of its new host. Genetic changes in the pathogen canarise either through ‘coincidental’ processes such as neutral drift orcoevolution of the pathogen and its reservoir host, or throughadaptive evolution of the pathogen during chains of transmission inhumans. Genetic changes of the new host might be more likely fordomesticated or endangered species than for humans.

Here we show that factors, such as ecological changes, thatincrease the R0 value of the potential pathogen to a level notsufficient to cause an epidemic (that is, R0 remains less than one)can greatly increase the length of the stochastic chains of diseasetransmission. These long transmission chains provide an opportu-nity for the pathogen to adapt to human hosts, and thus for thedisease to emerge.

Our model is illustrated in Fig. 1. Introductions occur stochas-tically from the natural reservoir of the pathogen, and each primarycase is followed by stochastic transmission that generates a variablenumber of subsequent infections in the human population. We

assume that the number of secondary cases follows a Poissondistribution with a mean equal to R0. Each introduction thusforms branched chains of transmission, which stutter to extinctionif R0 , 1, and the pathogen cannot evolve (m ¼ 0). The probabilitythat the pathogen evolves and a secondary infection is caused by themutant is equal to m for each of the secondary infections (we notethat m incorporates not only the pathogen’s mutation rate, but alsoits dynamics within the host and its transmissibility). We use multi-type branching processes6–9 to describe the initial spread of theinfection, incorporating the evolution of the pathogen.In the simplest case (see Fig. 2a) only onemutation is required for

the R0 of the evolved pathogen to exceed one. The probability that asingle introduction evolves, causing one (or more) infections withthe evolved pathogen (a filled circle in Fig. 1) before it goes extinct,depends very strongly on the R0 of the introduced pathogen,particularly for low m values as R0 approaches 1. This probabilityis approximately linearly dependent on the rate of evolution m.The probability that the introduction leads to an epidemic (the

‘probability of emergence’) depends on the probability of evolutionand the probability that the evolved infections do not go extinct dueto stochastic effects. In Fig. 2b we plot the probability of emergencefor three different R0 values of the evolved strain. The probability ofemergence approaches the probability of evolution when R0 of theevolved strain is large, and is lower when the R0 of the evolved strain

Figure 1 Schematic for the emergence of an infectious disease. Introductions from the

reservoir are followed by chains of transmission in the human population. Infections with

the introduced strain (open circles) have a basic reproductive number R 0 , 1. Pathogen

evolution generates an evolved strain (filled circles) with R 0 . 1. The infections caused

by the evolved strain can go on to cause an epidemic. Daggers indicate no further

transmission.

Figure 2 One-step evolution. A single change is required for the pathogen to evolve to

R 0 . 1. a, The probability that an introduction leads to an infection with an evolved strainof the pathogen (filled circle in Fig. 1) is highly sensitive to R 0, and is approximately

linearly dependent on the mutation rate m. Lines correspond to numerical solutions to the

branching process model (see Supplementary Information) and symbols correspond to

Monte-Carlo simulations following 105 introductions. b, The probability of emergence perintroduction depends on the R 0 value of the introduced pathogen and of the evolved

pathogen. The solid, dashed and dotted lines correspond to the evolved pathogen having

an R 0 of 1,000, 1.5 and 1.2 respectively.

letters to nature

NATURE |VOL 426 | 11 DECEMBER 2003 | www.nature.com/nature 659© 2003 Nature Publishing Group

is close to 1.We find that the probability of emergence dependsmoststrongly on the R0 of the introduced pathogen, increases approxi-mately linearly with themutation rate m, and depends onlymodestlyon the R0 of the evolved pathogen.We extend the simple one-step mutation model to consider the

situation in which multiple evolutionary changes are required forthe pathogen to attain R0 . 1 in the human population. We beginwith a simple scenario, which we call the jackpot model, where theR0 of the pathogen with the intermediate mutations is the same asthat of the introduced pathogen, and where only the addition of thefinal mutation results in an increase in R0 to greater than one. Asseen in Fig. 3, increasing the number of required evolutionary stepsgreatly reduces the probability of emergence and increases itssensitivity to changes in R 0. The probability of emergence isapproximately proportional to the mutation rate to the power ofthe number of evolutionary steps required (see SupplementaryInformation).The fitness landscape on which evolution occurs is important in

determining the outcome10. Figure 3b illustrates this for the case of asingle intermediate type. As should be expected, changing thejackpot model to an additive model, where the fitness of theintermediate is the average of the fitness of the introduced strainand fully evolved strain, increases the probability that the pathogenevolves to R0 . 1, whereas changing it to a fitness valley model,where the fitness of the intermediate is lower than the fitness of theintroduced strain, decreases the probability of emergence.The key characteristic distinguishing our model from the con-

ventional view is the R 0 of the introduced pathogen. In theconventional view the R0 of these infections must be greater thanone, whereas in the mechanism described here it is less than one,and evolution during the stochastic chains of transmission allowsR0 to increase above one. In the case of human infections suchas HIV1,11,12, SARS13–17 and (potentially) monkeypox18–21, sero-prevalence studies among groups at high risk of infection fromthe reservoir would allow evaluation of whether crossover events areusually dead ends, as we expect in our model, or whether they areassociated with a large number of secondary cases. In the case ofdiseases emerging into non-human populations, such as the Nipahvirus, which moved from bats to pigs22–24, it may be possible toconduct additional tests involving controlled experimental infec-tions to estimate the R0 (in this case of the bat Nipah virus in pigs)and to determine whether it evolves during the course of chains oftransmission. Such studies may additionally help to identify patho-gens that have an ability to evolve rapidly and thus have a highpotential for emergence.

The framework presented here has special relevance for patho-gens that have been driven to extinction by vaccination. In the caseof smallpox there are probably reservoirs of related zoonoses (suchas, but not restricted to, monkeypox) from which smallpox mayhave originated. Although the R0 of monkeypox in the humanpopulation is clearly less than one, there are occasional chains oftransmission in the human population18–21. As the level of herdimmunity to smallpox wanes in the absence of continued vacci-nation, we expect an increase in R0 of infections with monkeypoxalbeit to a level still less than one (the smallpox vaccine providesabout 85% cross immunity against monkeypox). Our resultssuggest that this increase in the effective R0 of monkeypox in thehuman population could markedly increase the probability ofevolution of monkeypox, allowing it to emerge into a successfulhuman pathogen (which, depending on the evolutionary trajectoryfollowed, may be similar to or differ from smallpox).

The present study could be extended in a number of directions.These include explicitly incorporating the details of ecologicalinteractions such as heterogeneity in the transmission in differentareas and subpopulations2,25,26,27 and incorporating genetic diversityof the pathogen in its reservoir. Finally, we note that this frameworkcan be applied to the more general problem of biological inva-sions28,29. A

MethodsWe describe the dynamics and evolution of emerging diseases as a multi-type branchingprocess with the following probability-generating functions6–9:

f iðs1; s2;…; smÞ ¼ exp½2ð12 mÞRðiÞ0 ð12 siÞ%exp½2mRðiÞ

0 ð12 siþ1Þ%; i¼ 1;…;m2 1

f mðs1; s2;…; smÞ ¼ exp½2RðmÞ0 ð12 smÞ%

We calculated the extinction probabilities of the above process numerically. For detailsand definitions see Supplementary Information.

Received 6 July; accepted 25 September 2003; doi:10.1038/nature02104.

1. Hahn, B. H., Shaw, G. M., De Cock, K. M. & Sharp, P. M. AIDS as a zoonosis: scientific and public

health implications. Science 287, 607–614 (2000).

2. May, R. M., Gupta, S. & McLean, A. R. Infectious disease dynamics: What characterizes a successful

invader? Phil. Trans. R. Soc. Lond. B 356, 901–910 (2001).

3. Anderson, R. & May, R. Infectious Diseases of Humans: Dynamics and Control 1st edn (Oxford Univ.

Press, Oxford, 1991).

4. Schrag, S. & Wiener, P. Emerging infectious disease: what are the relative roles of ecology and

evolution? Trends Ecol. Evol. 10, 319–324 (1995).

5. Wallace, B. Can “stepping stones” form stairways? Am. Nat. 133, 578–579 (1989).

6. Athreya, K. B. & Ney, P. Branching Processes (Springer, Berlin/New York, 1972).

7. Demetrius, L., Schuster, P. & Sigmund, K. Polynucleotide evolution and branching processes. Bull.

Math. Biol. 47, 239–262 (1985).

8. Stivers, D. N., Kimmel, M. & Axelrod, D. E. A discrete-time, multi-type generational inheritance

branching process model of cell proliferation. Math. Biosci. 137, 25–50 (1996).

9. Wilke, C. O. Probability of fixation of an advantageous mutant in a viral quasispecies. Genetics 163,

467–474 (2003).

10. Wright, S. The role of mutation, inbreeding, crossbreeding and selection in evolution. Proc. 6th Int.

Cong. Genet. 1, 356–366 (1932).

11. Holmes, E. C. On the origin and evolution of the human immunodeficiency virus (HIV). Biol. Rev.

Figure 3 Multiple-step evolution. Here multiple evolutionary changes are required forevolution of the pathogen to have an R 0 . 1. a, Jackpot model with m ¼ 0.1 and n

intermediate changes each with R 0 equal to that of the introduced pathogen: increasing

the number of steps (n) greatly decreases the probability of evolution, and makes it more

sensitive to the R 0 of the introduced pathogen. b, Alternative multi-step models for theone-intermediate (n ¼ 1) case. The jackpot model (solid line), additive model (dashed

line) and fitness valley model (dotted line) are shown (see text for details).

letters to nature

NATURE | VOL 426 | 11 DECEMBER 2003 | www.nature.com/nature660 © 2003 Nature Publishing Group

is close to 1.We find that the probability of emergence dependsmoststrongly on the R0 of the introduced pathogen, increases approxi-mately linearly with themutation rate m, and depends onlymodestlyon the R0 of the evolved pathogen.We extend the simple one-step mutation model to consider the

situation in which multiple evolutionary changes are required forthe pathogen to attain R0 . 1 in the human population. We beginwith a simple scenario, which we call the jackpot model, where theR0 of the pathogen with the intermediate mutations is the same asthat of the introduced pathogen, and where only the addition of thefinal mutation results in an increase in R0 to greater than one. Asseen in Fig. 3, increasing the number of required evolutionary stepsgreatly reduces the probability of emergence and increases itssensitivity to changes in R 0. The probability of emergence isapproximately proportional to the mutation rate to the power ofthe number of evolutionary steps required (see SupplementaryInformation).The fitness landscape on which evolution occurs is important in

determining the outcome10. Figure 3b illustrates this for the case of asingle intermediate type. As should be expected, changing thejackpot model to an additive model, where the fitness of theintermediate is the average of the fitness of the introduced strainand fully evolved strain, increases the probability that the pathogenevolves to R0 . 1, whereas changing it to a fitness valley model,where the fitness of the intermediate is lower than the fitness of theintroduced strain, decreases the probability of emergence.The key characteristic distinguishing our model from the con-

ventional view is the R 0 of the introduced pathogen. In theconventional view the R0 of these infections must be greater thanone, whereas in the mechanism described here it is less than one,and evolution during the stochastic chains of transmission allowsR0 to increase above one. In the case of human infections suchas HIV1,11,12, SARS13–17 and (potentially) monkeypox18–21, sero-prevalence studies among groups at high risk of infection fromthe reservoir would allow evaluation of whether crossover events areusually dead ends, as we expect in our model, or whether they areassociated with a large number of secondary cases. In the case ofdiseases emerging into non-human populations, such as the Nipahvirus, which moved from bats to pigs22–24, it may be possible toconduct additional tests involving controlled experimental infec-tions to estimate the R0 (in this case of the bat Nipah virus in pigs)and to determine whether it evolves during the course of chains oftransmission. Such studies may additionally help to identify patho-gens that have an ability to evolve rapidly and thus have a highpotential for emergence.

The framework presented here has special relevance for patho-gens that have been driven to extinction by vaccination. In the caseof smallpox there are probably reservoirs of related zoonoses (suchas, but not restricted to, monkeypox) from which smallpox mayhave originated. Although the R0 of monkeypox in the humanpopulation is clearly less than one, there are occasional chains oftransmission in the human population18–21. As the level of herdimmunity to smallpox wanes in the absence of continued vacci-nation, we expect an increase in R0 of infections with monkeypoxalbeit to a level still less than one (the smallpox vaccine providesabout 85% cross immunity against monkeypox). Our resultssuggest that this increase in the effective R0 of monkeypox in thehuman population could markedly increase the probability ofevolution of monkeypox, allowing it to emerge into a successfulhuman pathogen (which, depending on the evolutionary trajectoryfollowed, may be similar to or differ from smallpox).

The present study could be extended in a number of directions.These include explicitly incorporating the details of ecologicalinteractions such as heterogeneity in the transmission in differentareas and subpopulations2,25,26,27 and incorporating genetic diversityof the pathogen in its reservoir. Finally, we note that this frameworkcan be applied to the more general problem of biological inva-sions28,29. A

MethodsWe describe the dynamics and evolution of emerging diseases as a multi-type branchingprocess with the following probability-generating functions6–9:

f iðs1; s2;…; smÞ ¼ exp½2ð12 mÞRðiÞ0 ð12 siÞ%exp½2mRðiÞ

0 ð12 siþ1Þ%; i¼ 1;…;m2 1

f mðs1; s2;…; smÞ ¼ exp½2RðmÞ0 ð12 smÞ%

We calculated the extinction probabilities of the above process numerically. For detailsand definitions see Supplementary Information.

Received 6 July; accepted 25 September 2003; doi:10.1038/nature02104.

1. Hahn, B. H., Shaw, G. M., De Cock, K. M. & Sharp, P. M. AIDS as a zoonosis: scientific and public

health implications. Science 287, 607–614 (2000).

2. May, R. M., Gupta, S. & McLean, A. R. Infectious disease dynamics: What characterizes a successful

invader? Phil. Trans. R. Soc. Lond. B 356, 901–910 (2001).

3. Anderson, R. & May, R. Infectious Diseases of Humans: Dynamics and Control 1st edn (Oxford Univ.

Press, Oxford, 1991).

4. Schrag, S. & Wiener, P. Emerging infectious disease: what are the relative roles of ecology and

evolution? Trends Ecol. Evol. 10, 319–324 (1995).

5. Wallace, B. Can “stepping stones” form stairways? Am. Nat. 133, 578–579 (1989).

6. Athreya, K. B. & Ney, P. Branching Processes (Springer, Berlin/New York, 1972).

7. Demetrius, L., Schuster, P. & Sigmund, K. Polynucleotide evolution and branching processes. Bull.

Math. Biol. 47, 239–262 (1985).

8. Stivers, D. N., Kimmel, M. & Axelrod, D. E. A discrete-time, multi-type generational inheritance

branching process model of cell proliferation. Math. Biosci. 137, 25–50 (1996).

9. Wilke, C. O. Probability of fixation of an advantageous mutant in a viral quasispecies. Genetics 163,

467–474 (2003).

10. Wright, S. The role of mutation, inbreeding, crossbreeding and selection in evolution. Proc. 6th Int.

Cong. Genet. 1, 356–366 (1932).

11. Holmes, E. C. On the origin and evolution of the human immunodeficiency virus (HIV). Biol. Rev.

Figure 3 Multiple-step evolution. Here multiple evolutionary changes are required forevolution of the pathogen to have an R 0 . 1. a, Jackpot model with m ¼ 0.1 and n

intermediate changes each with R 0 equal to that of the introduced pathogen: increasing

the number of steps (n) greatly decreases the probability of evolution, and makes it more

sensitive to the R 0 of the introduced pathogen. b, Alternative multi-step models for theone-intermediate (n ¼ 1) case. The jackpot model (solid line), additive model (dashed

line) and fitness valley model (dotted line) are shown (see text for details).

letters to nature

NATURE | VOL 426 | 11 DECEMBER 2003 | www.nature.com/nature660 © 2003 Nature Publishing Group

Natural selection: infection begets transmission, transmission begets epidemics

General theory

See also, Lloyd-Smith et al Epidemics 2014 and Park et al Proc R Soc B 2013

Page 9: Session 4: Modeling of potential pandemics

Predicting pandemics: empiricism meets theory

• Infection predicts transmission.

• Transmission predicts epidemic emergence.

• Focus on infections that cause human cases and clusters → Surveillance

Page 10: Session 4: Modeling of potential pandemics

Surveillance & response• Surveillance must be prompt.

• Human infections & clusters are alone cause for enhanced response.

• Most timely response would be based on timely reporting of cases.

968

The ongoing H5N1 influenza epidemicin Asian bird populations poses risksto public as well as animal health (1),

due to the potential for cross-species trans-mission to humans and subsequent reas-sortment of avian and human influenza

viruses in coinfect-ed individuals (2).Reassortment couldlink the high trans-missibility associat-

ed with human-adapted viruses with thehigh rates of mortality observed in the cur-rent human cases and thus trigger a poten-tially devastating pandemic.

Good pandemic planning (3) and world-wide surveillance is central in mounting aneffective global response to combat suchthreats. However, surveillance must be linkedto appropriate analysis for the stuttering be-ginnings of a threatening epidemic to be de-tected rapidly and reliably. Currently theWorld Health Organization (WHO) tracksthe number of avian-to-human and possiblehuman-to-human transmission events report-ed. Detection of even a single case of human-to-human transmission of an avian virus cancause an increase in pandemic alert levels—at potentially high cost to affected countries(particularly if travel advisories are issued).Here, we argue that observation of low-levelhuman-to-human transmission is not a keythreshold, but rather the primary focus shouldbe on detecting increases in viral transmissi-bility at a stage where containment might befeasible. To help ongoing consultation (4), wepresent a method to detect increases in viraltransmissibility based on examination ofclusters of human cases.

The WHO Global Influenza Network(5) was established in 1948, and todaycomprises 4 collaborating centers and 112institutions recognized as WHO national

influenza centers in 83 countries. The ma-jor focus of the network is on compiling in-formation for influenza vaccine formula-tion, based on the analysis of viral isolatescollected in the participating countries. Italso serves as a mechanism for alertingcountries to the emergence of strains withpandemic potential or unusual pathogenic-ity. A survey of the network in 2002 result-ed in the development of a WHO actionplan for the next 5 years to improve on cur-rent surveillance weaknesses (6). Issueshighlighted were the current wide variabil-ity in the number of viral isolates submit-

ted by different countries, and the limitedfraction of isolates that undergo detailedanalysis and sequencing.

Surveillance of avian and livestock (par-ticularly porcine) influenza is less satisfac-tory. Government veterinary services are re-sponsible for surveillance, with the Foodand Agriculture Organization (FAO) and theWorld Organization for Animal Health(OIE) collating data from those countriessubmitting information on disease out-breaks. There is no systematic surveillancefor influenza in livestock or wild birds basedon random sampling or other well-definedsampling schemes. Most data are collatedfrom outbreaks causing high morbidity andmortality. Furthermore, outside of the de-veloped world, few countries have the vet-erinary services infrastructure to undertakeeffective surveillance. As the current H5N1avian epidemic highlights, the zoonotic ori-gin of pandemic influenza strains meansthat improved surveillance and control ofanimal (particularly avian) influenza needsto be a priority—as much to safeguard pub-lic health as to promote animal welfare.

The demands of annual vaccine prepa-ration inevitably mean that greatest empha-sis is placed on monitoring antigenic vari-ation and understanding its molecular ba-sis. However, improving pandemic surveil-lance additionally requires more research(in secure laboratory settings) to measurethe frequency of viral reassortment in hostsexposed to multiple different viral strainsand to define the detailed genetic basis ofhost specificity, pathogenicity, and trans-missibility (7).

How do we quantify the possible risks ofa pandemic strain’s emerging, and how canwe rapidly but reliably detect the emer-gence of such a strain? Simple mathemati-cal analysis can provide some insights. Atany time point, the risk of a reassortmentevent is proportional to the number of peo-ple coinfected with human and avianstrains. Making the reasonable assumptionsthat 10% of the population are infected withhuman influenza over the 12 weeks of atypical influenza season (8), and that thereis a 1-day window in early infection wherecoinfection with an avian strain is possible,then 0.12% of the population are suscepti-ble to coinfection with an avian strain atany one time. Hence, even if reassortmentis certain following coinfection, the proba-bility of a reassortment event having oc-curred after n cases of avian influenza inhumans is 1 – (1 – 0.0012)n; so 600 humaninfections would be required for a 50%chance of reassortment, and around 45 for a5% chance. If reassortment is a rare out-

P U B L I C H E A LT H

Public Health Risk from theAvian H5N1 Influenza Epidemic

Neil M. Ferguson,*† Christophe Fraser,†Christl A. Donnelly, Azra C. Ghani, Roy M. Anderson

POLICY FORUM

The authors are in the Department of InfectiousDisease Epidemiology, Faculty of Medicine, ImperialCollege London, St. Mary’s campus, Norfolk Place,London W2 1PG, UK.

*Author for correspondence. E-mail: [email protected]†These authors contributed equally.

1625% (R0 ~ 0.33)

15% (R0 ~ 0.2)

5% (R0 ~ 0.05)

141210

86420

0.80.60.40.2

00 10 20

20 40 60 80 100 120 140 160 180 200

30 40Cumulative number

of avian to human cases

Thre

shol

dR0

Cumulative numberof avian to human cases

Analyzing avian influenza surveillance data.(Top) Illustration of the reliability of R0 estima-tion methods. R0 estimates (the shading delin-eates the 95% confidence intervals) are shownderived from data on the number of avian-to-human cases seen, the number of human-hu-man clusters, and the size of the largest cluster.The surveillance data used was generated bysimulating an avian influenza epidemic with R0

= 0.2 in humans. (Bottom) Threshold size of thelargest cluster expected by chance for a range oflevels of human-to-human transmission, asquantified by the proportion of avian-to-humancases generating secondary cases (approximateR0 values are also shown). Anomalous behaviormight be suspected if a cluster exceeds thisthreshold size. Note how the expected maxi-mum cluster size increases cases accumulate.See supporting online material for methods.

14 MAY 2004 VOL 304 SCIENCE www.sciencemag.org

Enhanced online atwww.sciencemag.org/cgi/content/full/304/5673/968

968

The ongoing H5N1 influenza epidemicin Asian bird populations poses risksto public as well as animal health (1),

due to the potential for cross-species trans-mission to humans and subsequent reas-sortment of avian and human influenza

viruses in coinfect-ed individuals (2).Reassortment couldlink the high trans-missibility associat-

ed with human-adapted viruses with thehigh rates of mortality observed in the cur-rent human cases and thus trigger a poten-tially devastating pandemic.

Good pandemic planning (3) and world-wide surveillance is central in mounting aneffective global response to combat suchthreats. However, surveillance must be linkedto appropriate analysis for the stuttering be-ginnings of a threatening epidemic to be de-tected rapidly and reliably. Currently theWorld Health Organization (WHO) tracksthe number of avian-to-human and possiblehuman-to-human transmission events report-ed. Detection of even a single case of human-to-human transmission of an avian virus cancause an increase in pandemic alert levels—at potentially high cost to affected countries(particularly if travel advisories are issued).Here, we argue that observation of low-levelhuman-to-human transmission is not a keythreshold, but rather the primary focus shouldbe on detecting increases in viral transmissi-bility at a stage where containment might befeasible. To help ongoing consultation (4), wepresent a method to detect increases in viraltransmissibility based on examination ofclusters of human cases.

The WHO Global Influenza Network(5) was established in 1948, and todaycomprises 4 collaborating centers and 112institutions recognized as WHO national

influenza centers in 83 countries. The ma-jor focus of the network is on compiling in-formation for influenza vaccine formula-tion, based on the analysis of viral isolatescollected in the participating countries. Italso serves as a mechanism for alertingcountries to the emergence of strains withpandemic potential or unusual pathogenic-ity. A survey of the network in 2002 result-ed in the development of a WHO actionplan for the next 5 years to improve on cur-rent surveillance weaknesses (6). Issueshighlighted were the current wide variabil-ity in the number of viral isolates submit-

ted by different countries, and the limitedfraction of isolates that undergo detailedanalysis and sequencing.

Surveillance of avian and livestock (par-ticularly porcine) influenza is less satisfac-tory. Government veterinary services are re-sponsible for surveillance, with the Foodand Agriculture Organization (FAO) and theWorld Organization for Animal Health(OIE) collating data from those countriessubmitting information on disease out-breaks. There is no systematic surveillancefor influenza in livestock or wild birds basedon random sampling or other well-definedsampling schemes. Most data are collatedfrom outbreaks causing high morbidity andmortality. Furthermore, outside of the de-veloped world, few countries have the vet-erinary services infrastructure to undertakeeffective surveillance. As the current H5N1avian epidemic highlights, the zoonotic ori-gin of pandemic influenza strains meansthat improved surveillance and control ofanimal (particularly avian) influenza needsto be a priority—as much to safeguard pub-lic health as to promote animal welfare.

The demands of annual vaccine prepa-ration inevitably mean that greatest empha-sis is placed on monitoring antigenic vari-ation and understanding its molecular ba-sis. However, improving pandemic surveil-lance additionally requires more research(in secure laboratory settings) to measurethe frequency of viral reassortment in hostsexposed to multiple different viral strainsand to define the detailed genetic basis ofhost specificity, pathogenicity, and trans-missibility (7).

How do we quantify the possible risks ofa pandemic strain’s emerging, and how canwe rapidly but reliably detect the emer-gence of such a strain? Simple mathemati-cal analysis can provide some insights. Atany time point, the risk of a reassortmentevent is proportional to the number of peo-ple coinfected with human and avianstrains. Making the reasonable assumptionsthat 10% of the population are infected withhuman influenza over the 12 weeks of atypical influenza season (8), and that thereis a 1-day window in early infection wherecoinfection with an avian strain is possible,then 0.12% of the population are suscepti-ble to coinfection with an avian strain atany one time. Hence, even if reassortmentis certain following coinfection, the proba-bility of a reassortment event having oc-curred after n cases of avian influenza inhumans is 1 – (1 – 0.0012)n; so 600 humaninfections would be required for a 50%chance of reassortment, and around 45 for a5% chance. If reassortment is a rare out-

P U B L I C H E A LT H

Public Health Risk from theAvian H5N1 Influenza Epidemic

Neil M. Ferguson,*† Christophe Fraser,†Christl A. Donnelly, Azra C. Ghani, Roy M. Anderson

POLICY FORUM

The authors are in the Department of InfectiousDisease Epidemiology, Faculty of Medicine, ImperialCollege London, St. Mary’s campus, Norfolk Place,London W2 1PG, UK.

*Author for correspondence. E-mail: [email protected]†These authors contributed equally.

1625% (R0 ~ 0.33)

15% (R0 ~ 0.2)

5% (R0 ~ 0.05)

141210

86420

0.80.60.40.2

00 10 20

20 40 60 80 100 120 140 160 180 200

30 40Cumulative number

of avian to human cases

Thre

shol

dR0

Cumulative numberof avian to human cases

Analyzing avian influenza surveillance data.(Top) Illustration of the reliability of R0 estima-tion methods. R0 estimates (the shading delin-eates the 95% confidence intervals) are shownderived from data on the number of avian-to-human cases seen, the number of human-hu-man clusters, and the size of the largest cluster.The surveillance data used was generated bysimulating an avian influenza epidemic with R0

= 0.2 in humans. (Bottom) Threshold size of thelargest cluster expected by chance for a range oflevels of human-to-human transmission, asquantified by the proportion of avian-to-humancases generating secondary cases (approximateR0 values are also shown). Anomalous behaviormight be suspected if a cluster exceeds thisthreshold size. Note how the expected maxi-mum cluster size increases cases accumulate.See supporting online material for methods.

14 MAY 2004 VOL 304 SCIENCE www.sciencemag.org

Enhanced online atwww.sciencemag.org/cgi/content/full/304/5673/968

968

The ongoing H5N1 influenza epidemicin Asian bird populations poses risksto public as well as animal health (1),

due to the potential for cross-species trans-mission to humans and subsequent reas-sortment of avian and human influenza

viruses in coinfect-ed individuals (2).Reassortment couldlink the high trans-missibility associat-

ed with human-adapted viruses with thehigh rates of mortality observed in the cur-rent human cases and thus trigger a poten-tially devastating pandemic.

Good pandemic planning (3) and world-wide surveillance is central in mounting aneffective global response to combat suchthreats. However, surveillance must be linkedto appropriate analysis for the stuttering be-ginnings of a threatening epidemic to be de-tected rapidly and reliably. Currently theWorld Health Organization (WHO) tracksthe number of avian-to-human and possiblehuman-to-human transmission events report-ed. Detection of even a single case of human-to-human transmission of an avian virus cancause an increase in pandemic alert levels—at potentially high cost to affected countries(particularly if travel advisories are issued).Here, we argue that observation of low-levelhuman-to-human transmission is not a keythreshold, but rather the primary focus shouldbe on detecting increases in viral transmissi-bility at a stage where containment might befeasible. To help ongoing consultation (4), wepresent a method to detect increases in viraltransmissibility based on examination ofclusters of human cases.

The WHO Global Influenza Network(5) was established in 1948, and todaycomprises 4 collaborating centers and 112institutions recognized as WHO national

influenza centers in 83 countries. The ma-jor focus of the network is on compiling in-formation for influenza vaccine formula-tion, based on the analysis of viral isolatescollected in the participating countries. Italso serves as a mechanism for alertingcountries to the emergence of strains withpandemic potential or unusual pathogenic-ity. A survey of the network in 2002 result-ed in the development of a WHO actionplan for the next 5 years to improve on cur-rent surveillance weaknesses (6). Issueshighlighted were the current wide variabil-ity in the number of viral isolates submit-

ted by different countries, and the limitedfraction of isolates that undergo detailedanalysis and sequencing.

Surveillance of avian and livestock (par-ticularly porcine) influenza is less satisfac-tory. Government veterinary services are re-sponsible for surveillance, with the Foodand Agriculture Organization (FAO) and theWorld Organization for Animal Health(OIE) collating data from those countriessubmitting information on disease out-breaks. There is no systematic surveillancefor influenza in livestock or wild birds basedon random sampling or other well-definedsampling schemes. Most data are collatedfrom outbreaks causing high morbidity andmortality. Furthermore, outside of the de-veloped world, few countries have the vet-erinary services infrastructure to undertakeeffective surveillance. As the current H5N1avian epidemic highlights, the zoonotic ori-gin of pandemic influenza strains meansthat improved surveillance and control ofanimal (particularly avian) influenza needsto be a priority—as much to safeguard pub-lic health as to promote animal welfare.

The demands of annual vaccine prepa-ration inevitably mean that greatest empha-sis is placed on monitoring antigenic vari-ation and understanding its molecular ba-sis. However, improving pandemic surveil-lance additionally requires more research(in secure laboratory settings) to measurethe frequency of viral reassortment in hostsexposed to multiple different viral strainsand to define the detailed genetic basis ofhost specificity, pathogenicity, and trans-missibility (7).

How do we quantify the possible risks ofa pandemic strain’s emerging, and how canwe rapidly but reliably detect the emer-gence of such a strain? Simple mathemati-cal analysis can provide some insights. Atany time point, the risk of a reassortmentevent is proportional to the number of peo-ple coinfected with human and avianstrains. Making the reasonable assumptionsthat 10% of the population are infected withhuman influenza over the 12 weeks of atypical influenza season (8), and that thereis a 1-day window in early infection wherecoinfection with an avian strain is possible,then 0.12% of the population are suscepti-ble to coinfection with an avian strain atany one time. Hence, even if reassortmentis certain following coinfection, the proba-bility of a reassortment event having oc-curred after n cases of avian influenza inhumans is 1 – (1 – 0.0012)n; so 600 humaninfections would be required for a 50%chance of reassortment, and around 45 for a5% chance. If reassortment is a rare out-

P U B L I C H E A LT H

Public Health Risk from theAvian H5N1 Influenza Epidemic

Neil M. Ferguson,*† Christophe Fraser,†Christl A. Donnelly, Azra C. Ghani, Roy M. Anderson

POLICY FORUM

The authors are in the Department of InfectiousDisease Epidemiology, Faculty of Medicine, ImperialCollege London, St. Mary’s campus, Norfolk Place,London W2 1PG, UK.

*Author for correspondence. E-mail: [email protected]†These authors contributed equally.

1625% (R0 ~ 0.33)

15% (R0 ~ 0.2)

5% (R0 ~ 0.05)

141210

86420

0.80.60.40.2

00 10 20

20 40 60 80 100 120 140 160 180 200

30 40Cumulative number

of avian to human cases

Thre

shol

dR0

Cumulative numberof avian to human cases

Analyzing avian influenza surveillance data.(Top) Illustration of the reliability of R0 estima-tion methods. R0 estimates (the shading delin-eates the 95% confidence intervals) are shownderived from data on the number of avian-to-human cases seen, the number of human-hu-man clusters, and the size of the largest cluster.The surveillance data used was generated bysimulating an avian influenza epidemic with R0

= 0.2 in humans. (Bottom) Threshold size of thelargest cluster expected by chance for a range oflevels of human-to-human transmission, asquantified by the proportion of avian-to-humancases generating secondary cases (approximateR0 values are also shown). Anomalous behaviormight be suspected if a cluster exceeds thisthreshold size. Note how the expected maxi-mum cluster size increases cases accumulate.See supporting online material for methods.

14 MAY 2004 VOL 304 SCIENCE www.sciencemag.org

Enhanced online atwww.sciencemag.org/cgi/content/full/304/5673/968

Page 11: Session 4: Modeling of potential pandemics

Pre-pandemic vaccine strain selection

• Timely development of vaccines could be transformative.

• Vaccine seed stocks can speed this up, but also focus on other rate limiting steps, esp. international agreements on regulation and conduct of human trials.

• Shouldn’t aims be:

• to prioritise strains with evidence of infection and transmission?

• to cover antigen space, and monitor antigenic drift?

• to plug gaps in surveillance?

• to make more / faster seed stocks (Dormitzer et al. STM 2013)?

Page 12: Session 4: Modeling of potential pandemics

Conclusions• Direct benefits for enhanced surveillance and model-

based prediction of GOF experiment with PPP should not be overstated.

• Indirect benefits of basic science are likely huge, but rationale for work dangerous pathogens requires benefits that outweigh risks and opportunity costs.

• Benefits of GOF with PPP for pre-pandemic vaccine production should be probed in depth.

• Risks are real and present (Lipsitch & Inglesby mBio 2014).

Page 13: Session 4: Modeling of potential pandemics