Why is the world so poorly prepared for a pandemic of hypervirulent avian influenza?

Abstract

The world is now extremely poorly prepared to counter a possible pandemic of hypervirulent H5N1 influenza. Most countries are planning for nothing worse than the Spanish flu pandemic. It may be possible that this can in large measure be explained as a consequence of an epidemic of wishful thinking, which may already have infected the health authorities (and parts of the scientific community as well) in most countries in the world. However, it may also be possible that it can have happened as a consequence of too little contact between medical scientists and more general biologists (natural scientists) from disciplines such as ornithology, ecology and evolutionary biology. This may have led to a lack of proper understanding among medical scientists (and health bureaucrats) of the nature of evolutionary processes affecting influenza viruses, as regards the evolution of host species adaptation, infectivity and virulence properties, and also a lack of appreciation of the ways in which such forms of evolutionary adaptation depend on ecological boundary conditions that have radically changed, comparing the world in 2006 to the world in 1918. While the Spanish flu virus possibly might be compared to a one-headed monster, it may be possible that highly virulent varieties of H5N1 virus might better be compared to a three-headed one – because there is evidence of at least three independent virulence factors connected with three different genes. It is highly unlikely that all of the high-virulence alleles will simultaneously mutate and disappear if and when the haemagglutinin gene changes so as to make the haemagglutinin molecule better adapted for the human-type (alpha-2,6-linked) receptor (which is a necessary prerequisite in order that a pandemic with H5N1 virus may start). It is more probable that evolutionary adaptation of the haemagglutinin of H5N1 viruses to the human-type receptor will happen without any simultaneous change in those other genetic properties that now are important for explaining the exceptionally high virulence of certain strains of avian-adapted H5N1 influenza virus. The change of the haemagglutinin molecule from avian adaptation to human adaptation must be expected to act as an additional virulence factor because it will enhance the total number of cells that can be infected (per host organism), increase the total rate of virus replication and potentiate the effects of the other virulence factors already present. The monster will then have four heads, not three, and case fatality rates must be expected to become even higher than they have been until now, perhaps reaching as high as 98–99% (at least in poor countries with less than optimal nutrition).

• avian influenza
• virulence factors
• virulence evolution
• risk estimates
• emergency preparedness

Introduction

It has been shown in an earlier article in this journal that a pandemic with hypervirulent avian influenza represents an acute threat of extraordinary magnitude, as regards the possible loss of human lives [1]. In terms of relative and total losses of human lives, the consequences might in a realistic worst case scenario be comparable to those attending a full-scale nuclear war – if not even worse. [See Appendix 1 for more detailed information about some of the likely geophysical and medical consequences of a full-scale nuclear war.] But the likelihood per time unit (e.g. within 1 year) that it could happen seems to be much higher for a pandemic with hypervirulent influenza than used to be the case during most of the so-called Cold War (with the exception of the Cuban crisis) for the possible outbreak of full-scale nuclear war.

During the Cold War, billions and billions were spent on military defence – for more than one reason (including traditional power politics and ideological motives); one of the important motives was a strong desire to prevent what everybody could understand was a very severe threat to all of us: something that must be expected to make everybody a loser, even among the survivors on the ‘winning’ side, in the case that it should really happen. But one might ask the question, whether there is any rational reason not to deploy similar efforts and resources – economically and otherwise – to try to counter a non-military threat that could have equally if not even more disastrous consequences (regarding possible losses of human lives), but may be considered far more likely to happen.

But the political world is not always very rational. Using Norway as an example, military spending is apparently not considered an equally high priority today as used to be the case during the Cold War. But we are still using tens of billions of Norwegian kroner (NOK) for military purposes every year (about 30 billion NOK in the Norwegian state budget for 2006), while 50 million NOK has been allocated for international efforts to improve emergency preparedness for a possible H5N1 pandemic. A few hundred million NOK have been used domestically for the same purpose, much of it for buying anti-influenza drugs (including 1.4 million user doses of Tamiflu). Total spending (domestically and internationally) over 1 year on emergency preparedness for an eventual influenza pandemic has therefore been only about 1% (order of magnitude) of our military budget for 1 year. It is not part of our problem that we don't have the necessary financial capacity to do much more to improve our emergency preparedness for an influenza pandemic; the value of our so-called petroleum fund (which is a public pension fund financed by the Norwegian state's net income from the petroleum and gas sector) is now more than 1500 billion NOK. And it would create no immediate problems for our domestic economy (e.g. in the form of inflation), if some of this money was used to support international efforts to reduce the risk of the outbreak of a pandemic with hypervirulent influenza and of harm reduction in case it still should happen.

Governments and parliaments cannot make correct decisions on how to counter an extraordinary health-related security threat unless they have received the information needed for a realistic understanding of the threat scenario

It might possibly be useful for Norwegian politicians to ask themselves the question, what is going to happen to the petroleum fund if there should be a global pandemic considerably worse than the Spanish flu pandemic – something that better might be compared to the Black Death, if not even worse [1]. It is not necessary to be a Nobel laureate in economy to predict that the world economy then must be expected to suffer a shock far worse than what happened in 1929 (and which led the world into the great depression of the early 1930s). In an economic sense, it may be possible that the world would go back almost to the Dark Ages, similar to what may also be expected to happen after a full-scale nuclear war [2]. Big companies might either go bankrupt or only barely survive – which means that the shareholders in the same companies (which might include the Norwegian state through its petroleum fund) also must be expected to suffer very severe economic losses. Would it not be a wise thing to do for the shareholders, even if they were thinking only about the safety of their own investments (not taking into consideration the loss of human lives that would also be expected to attend a pandemic of hypervirulent influenza), to pay some form of commercial insurance premium in order to reduce the risk of a world economic collapse that might reduce the value of their shares to almost nothing? Would, say, 3% of the present value of the shares be too much to pay as an annual premium, if this is what would be needed for significant reduction (say, by 50%) of the risk that almost all of the invested money suddenly might be lost? It should be added that we are talking here about a risk which must be considered very high not only because of the extraordinary magnitude of the damage potential, but also because the probability that it could happen in the comparatively near future must be considered fairly high – with the risk being defined as the product of the damage potential of a certain kind of hypothetical future event and the probability that it could happen. Although mathematical calculation of this probability may not yet be possible (except perhaps from historical statistics regarding the time distribution of earlier pandemics, say, over the last 500 years), it is possible, with >50% probability, that it could happen within the next 5 years. This represents a conservative (i.e. possibly over-optimistic) guesstimate.

However, our politicians are probably not as irrational as might appear to be the case. If they are not acting in a rational way, i.e. by allocating as much money as might realistically be needed for averting this kind of disaster, and for substantial harm reduction in case it still should happen, there is only one plausible explanation for this, i.e. that they have never received the information needed for a realistic estimation of the magnitude and likelihood of the threat. A government or a parliament cannot make good and sensible decisions about either health policy or military questions unless it has enough good and reliable technical, scientific and/or intelligence information on which to base these decisions. The GIGO principle (Garbage In, Garbage Out) holds true not only for computer models, but also for the political process in governments and parliaments.

The virus of malignant wishful thinking

There is no reason to believe that medical scientists or health bureaucrats may deliberately keep back information needed for the governments and parliaments to take correct decisions (before it has become too late) in order to prepare their nations for something that more properly should be regarded as a national security threat of the first order – with a damage potential orders of magnitude more serious than might be considered realistically possible, for example, for some political terrorist group to achieve. While <10 000 American citizens were killed in the terrorist attacks on the 11 September, it is possible that an H5N1 pandemic might kill more than 100 million American citizens – which means a damage potential four orders of magnitude worse than what happened in the terrorist attacks on the World Trade Center and the Pentagon. Nor is there any reason to believe that either medical scientists or health bureaucrats may deliberately give wrong or misleading information to governments and parliaments about such deadly serious matters.

However, it is not only viral and bacterial infections that are contagious, and that over a short time interval can infect so many people that we would call it an epidemic or a pandemic. It is well known that something similar can also happen with certain psychological phenomena, both benign and more malignant ones. Exuberant joy among a group of supporters after a football victory is a benign example, while panic and ethnic, racial or religious prejudices (or even hatred) are well-known examples of something less benign, but otherwise similar in the sense that they often may occur mainly as a consequence of communication processes within groups of individuals (similar to what also happens with infectious diseases), rather than as the sum of several independent individual (but parallel) responses to the same environmental or social circumstances (as when smokers frequently develop lung cancer – because they have all been exposed to the same tobacco smoke mutagens, but not because lung cancer is an infective disease). There seems to be no good reason not to add wishful thinking to this list of psychological phenomena that may spread within a vulnerable target population (or subpopulation) in a way not at all dissimilar to the epidemic dispersal of highly infective viral diseases, such as influenza. Of course, as individuals we can easily be affected by wishful thinking also in a non-infective form (e.g. when gambling, or when falling in love with somebody not rewarding our attention). But we may do so even more easily (and not so easily reverse our position) when our peers also do the same, and we can all give one another moral support when choosing to believe that this is not something that we need to seriously fear.

In this case, however, it may be possible that the epidemic could have hit especially hard among some of those particular subpopulations that more than anybody else ought to have been protected and immune, not for their own sake, but for the survival of society as a whole. With their normal professional contacts now serving as the main mechanism of transmission of what could be considered as not only an extraordinarily infective, but also extraordinarily dangerous ‘psychological virus’.

We should not take it for granted that medical professionals (including medical scientists) are more immune to wishful thinking than medical non-professionals (including members of parliaments and governments) when it comes to threats that are not part of their ordinary professional training and experience at the same time as they are also exceptionally dreadful, representing an extraordinary menace not only to their ordinary patients, but also for themselves and their own families. One would expect that almost all medical professionals (perhaps with the exception of some few medical scientists and health bureaucrats) should be very well trained, emotionally and otherwise, in the encounter with human disease, suffering and death. But this may not necessarily be the case when it happens (or could happen) on this scale.

It is a legacy of the Cold War and a consequence of the development of modern weapons of mass destruction that military personnel must be professionally and psychologically prepared to deal with truly apocalyptic scenarios (cf. 2–21); for them this is not something unthinkable, and they have to make contingency plans as realistic as possible in case something so dreadful actually should happen. But a medical scientist may not necessarily think in the same way as a top general does; he is not psychologically prepared for it, and a pandemic with something as infective as ordinary influenza and >50% lethality might imply a horror scenario more terrible than he can bear – making him an extremely susceptible victim to the virus of wishful thinking.

A fundamental prerequisite for all kinds of emergency planning, whether dealing with natural disasters (e.g. earthquakes, hurricanes, tsunamis) or disasters that have an anthropogenic cause (including terrorist attacks and military attacks from other nations) is to have a realistic understanding of the nature and possible magnitude of the threat, as well as of the likelihood that it could happen. If an overly nationalistic, expansionist and aggressive dictator (e.g. Adolf Hitler during the period 1935–1939) is not taken seriously enough by other nations in the region, and if they also underestimate his military capacity and the quality of his weapons, they might easily risk not being well enough prepared when that day comes that the same dictator has given his orders for attack. The fate of Norway in 1940 is a good example.

Preparing for the war that happened in 1918

The same is no less true when planning how to encounter a possible pandemic with some aggressive form of avian influenza. But a recent assessment [22] of national pandemic influenza preparedness plans from 21 countries in Europe (mostly European Union countries, but also Bulgaria, Romania, Norway and Switzerland) shows that every country has been planning for a ‘war’ corresponding to that which happened in 1918 (i.e. the Spanish flu pandemic) and not for the war that is more likely to come. Neither the world, our means of communication nor the virus itself are the same now as they used to be in 1918; therefore what all countries in the world should expect and try to prepare for now is a pandemic with something much more virulent than the Spanish flu virus and spreading much faster than the latter, not only because of the much higher population density in the world today compared with the situation in 1918 [1], but also because of modern forms of mass communication, making it possible for the virus to ride on jet planes when in 1918 it could not ride on anything faster than railroads and passenger ships.

Most of the national plans that were assessed indicated preparation based on estimates of attack rates ranging between 15% and 50% of the population [22], while our model calculations [1] suggest a more than 90% attack rate (in the absence of vigorous counter-measures) to be a more realistic figure. And the estimates for anticipated death rates ranged (when considering all the national plans) from 14 to 1685 per 100 000 population [22]. It is apparently forgotten in all of these national plans (or not considered to be relevant or important) that the cumulative mortality for all registered human cases of H5N1 influenza until now has been more than 50%, i.e. more than 50 000 dead per 100 000 population, which is more than 28 times worse than the ‘worst case’ of the most pessimistic one among those 21 European countries that had their influenza pandemic preparedness plans assessed [22]. [The numbers for the cumulative mortality continually change as new cases continue to be reported, but a recent global status [23] gives a total of 113 deaths among 204 registered cases.]

Yet, it must be considered more probable than the opposite that a pandemic H5N1 virus will become even more lethal than now if (or when) it comes. This is because the same mutations that will be needed to make it infective enough from human to human, so that it can spread as a new human pandemic, also will mean higher effective haemagglutinin receptor density (i.e. number of receptor molecules per unit of tissue volume) and a larger total number of such receptors per individual (when not only the ‘avian-type’ alpha-2,3-linked receptors, but also the ‘human-type’ alpha-2,6-linked receptors become available for attachment of the virus prior to infection at the cellular level). It will enhance the total number of cells per organism that can be easily infected (because there are many cells that have only alpha-2,6-linked receptors, but not alpha-2,3-linked receptors), which will in turn lead to more rapid viral replication and faster progression of the disease, leading to enhancement of mortality [1].

Using military planning as an analogy, it might be considered no better to plan for a repetition tomorrow of the Spanish flu pandemic (and nothing worse than that), than if we assume, very hypothetically and contrafactually, that the Norwegian military during the days of the Cold War should have made contingency plans for an attack from the east, where it was assumed that the invading forces would be using weapons and means of transportation that were no better than those weapons and horses that Russian soldiers were using when fighting against German troops during World War I. However, the real enemy – if and when an H5N1 pandemic comes – should not be expected to come on horseback equipped with World War I style guns. It is more probable that he will drop parachute soldiers from planes flying much faster than a horse can run, that he will use attack helicopters, and that he will use weapons of mass destruction far more powerful than anything that was used during World War I.

Misuse of historical statistics

There is no scientific argument other than historical statistics to use the Spanish flu pandemic as a basis for predicting what might be considered a realistic worst case scenario. However, this should be considered a thin argument, even from a purely statistical point of view. If we have a sample where n=3 or n=4 (the number of influenza pandemics during the 20th century, depending on whether or not we count the return of H1N1 influenza in 1977 as one of the pandemics), and if we find much variation for a parameter we want to measure (the virulence characteristics of each pandemic virus) even in this small sample, we should not expect that it would show us the entire range of natural variation for the parameter concerned. So we should not assume that the Spanish flu pandemic was the worst of its kind, considering all influenza pandemics over the last 500 years.

May be it was not. A probable influenza epidemic, which broke out in Europe in 1556 and lasted on and off until 1560, is reported to have had serious demographic consequences on both sides of the Atlantic [24]. One estimate places die-off in England from the influenza at no less than 20% of the entire population, and comparable losses occurred elsewhere in Europe [24]. Whether the influenza outbreak of the 1550s was a genuinely global phenomenon cannot be said for sure, but Japanese records also mention an outbreak of ‘coughing violence’ in 1556 from which ‘very many died’ [24]. However, we have much less precise demographic data and death statistics for earlier influenza pandemics (from the 16th century onwards) than we have for the Spanish flu pandemic, and dark figures must be expected to be even higher for earlier influenza pandemics than they were in many African and Asian countries (and probably also in Russia) in 1918. So it is difficult to judge from the historical source material available whether or not it is correct to consider the Spanish flu pandemic as the worst example of its kind known from the last half-millennium of human history (even though the scarce historical source material available might suggest that the influenza epidemic of the 1550s could have been even worse).

However, even if the historical statistical argument used as a basis for national pandemic planning is assumed, hypothetically, to be valid in the sense that the Spanish flu pandemic really might be the worst historical example of its kind, this does not necessarily give us reason for being confident that the Spanish flu pandemic is a realistic worst case scenario today – the worst kind of war that possibly might affect us in a comparatively near future. The influenza virus has a very short memory (no hippocampus, no temporal lobes, no brain!), and it does not study historical archives the way medical historians and statisticians (or health bureaucrats) can do. But it is an RNA virus with fairly unfaithful replication, which means it has a high mutation rate, and is capable of rapid evolutionary change in response to important changes in various external circumstances, i.e. in important ecological boundary conditions determining both the direction and the tempo of evolutionary changes in the virus.

Evolution of virulence properties in influenza viruses

Change of virulence and infectivity (transmissibility) properties among influenza viruses should not be regarded as only random and therefore essentially unpredictable events. It may be more fruitful for our understanding to regard these processes as representing the outcome of evolutionary processes that do not depend only on random mutation and random viral recombination events (where different viral strains may recombine parts of their genomes so as to form something new). They must in large measure also depend on selection processes that are not at all random, but dependent on properties of the host ecosystem that will determine what could be the optimal choice for the pathogen (i.e. will help to optimize its evolutionary fitness). A well-adapted pathogen should not be so virulent that it kills the host before it has had the opportunity to spread to other hosts. But it must also be expected, as long as the host is not killed too early, that a more rapidly replicating genotype of the pathogen will have an advantage compared with other, more slowly replicating genotypes of the same pathogen species – even if more rapid replication also should imply a higher degree of virulence in relation to the host. Evolution must be expected to proceed towards some form of optimal trade-off between these conflicting considerations. But the actual position of this optimum can be highly sensitive to changes in host population parameters that affect the ease of dispersal of the pathogen from one host individual to another. In general it may be expected that a higher population density for the host species – enhancing the facility of dispersal of the pathogen from one individual to another – will shift the evolutionary optimum for the pathogen in the direction of more rapid replication and therefore higher virulence.

It must be considered a plausible hypothesis that it may have been precisely this kind of ecological change (in the direction of much higher local population density than before) that may have favoured the evolution of new super-virulent strains of avian influenza. Humans are now keeping large numbers of poultry and swine under very crowded conditions that may be considered almost ideal for the evolution of highly virulent superpathogens. There can be no doubt that these conditions must have changed very much over the last 100 years both in Europe and in Asia. Conditions are not the same with modern industrial farming methods as they used to be on the farms in France or in China in 1918. Not only is the total number of humans (and the average human population density) considerably greater now than it was in 1918; the number of domestic animals has increased even more, which is a consequence not only of human population growth, but also of the impressive economic growth that has affected not only Europe and North America, but also several countries in Asia [25], [26]. As long as these external conditions determining what would be the evolutionary optimum for the avian influenza virus are not changed, we should not expect that the direction of evolution for the latter will be reversed, making it go back to something less virulent.

We can thus be fairly certain that ecological boundary conditions may favour the evolutionary emergence of considerably more virulent strains today than used to be the case during the period from 1500 to 1918. This conclusion appears to be supported by much of what is known about global patterns of influenza virus infections in wild birds and about the ecology of virus/host species interactions in wild bird populations, compared to what happens in domestic bird populations [27]. It follows as a direct consequence that we should no longer rely on historical statistics to estimate what could be a realistic worst case scenario for future influenza pandemics. Agricultural ecosystems have changed so much (with the maximal number of bird individuals per farm and their local population density both being much higher now than used to be the case before) as to make historical comparisons invalid. What would have been an unprofitable overkill capacity for a particular viral strain in 1918 (because it would have killed its host too early for efficient transmission to the next host individual) might instead now make the same viral strain a winner in the Darwinian evolutionary race, given the ecological boundary conditions that determine the position of the fitness optimum for the virus (regarding infectivity versus virulence properties) in many chicken farms today.

It may be possible that the ecological boundary conditions even in old-fashioned agricultural ecosystems – as they may have been functioning for millennia up to about 1918 – could have been sufficiently changed, compared with natural ecosystems, as to favour the evolution of influenza virus strains that have been much more virulent than one would normally expect to find among wild bird populations. In stark contrast to contemporary human influenza H1N1 viruses, the 1918 pandemic virus has been found to have the ability to replicate in the absence of trypsin, to cause death in mice and embryonated chicken eggs, and to display a high-growth phenotype in human bronchial epithelial cells [28]. The non-dependence of the virus on trypsin for replication can be explained as a consequence of the introduction of basic amino acids into (or adjacent to) the cleavage site of the haemagglutinin precursor protein, which is a change that is also found in some of the highly virulent influenza virus strains that now are circulating in both domestic and wild bird populations. It makes the haemagglutinin precursor independent of trypsin for cleavage prior to uptake of the virus into the cells and therefore facilitates systemic virus replication, making it easier for the virus to attack other organs than the upper and lower airways [27–32]. An evolutionary hypothesis could be raised that possibly might help to explain the emergence of this important virulence character in the Spanish flu virus:

It might be speculated that the avian-adapted influenza virus strains that circulate among wild birds and possess a trypsin-dependent haemagglutinin precursor protein may use the trypsin that comes from the pancreas and is normally found in the intestinal lumen of the birds for activation (cleavage of the haemagglutinin precursor molecule) before the viruses can infect the cells of the intestinal mucosa. But it might also be hypothesized that there are normally only a few main cycles of virus replication (or perhaps only one) within the mucosa (similar to what is thought to happen during rotavirus infection in humans), when a bird is infected with a trypsin-dependent strain of influenza virus. After an intestinal epithelial cell has been infected, it might be hypothesized that it will shed a new generation of virus particles into the intestinal lumen, from which most of the virus particles will go into the cloaca (and from there out into the water where waterfowl are swimming before the virus particles can infect new host individuals when they drink the water). But this will happen without substantial re-infection of other cells in the intestinal mucosa, causing a new cycle of production of virus particles (i.e. a new generation of virus particles within the same infected individual).

If the haemagglutinin precursor protein becomes trypsin-independent, this may possibly facilitate re-infection (by viruses shed from the cohort of cells that were first infected) of other (previously uninfected) cells within the intestinal mucosa – because it is no longer necessary for the virus to be shed into the intestinal lumen for activation by trypsin. A much larger proportion of each new generation of virus particles can then be used for re-infection of new mucosal cells, rather than going to the cloaca and the external environment. This may increase the total virus production per animal per infectious episode (which means higher infectivity for the virus, i.e. enhanced probability that other individuals will be infected) at the same time as it also enhances the likelihood that an infected animal may die through enhanced virulence. We can speculate that this might be a form of evolutionary change that could enhance the Darwinian fitness of the influenza virus not only under such industrial farming conditions as are common in Europe, North America and several Asian countries today, but even in poultry populations that were living on old-fashioned farms in 1918 and before (may be even as far back in history as to the Neolithic farmer societies) – at the same time as this could be something that would not have been selected for (i.e. would not have enhanced the Darwinian fitness) in wild bird populations. [See Appendix 2 for information about a possible local outbreak of highly pathogenic avian influenza that is mentioned in the Bible. This episode is probably the oldest one of its kind yet known from recorded history (even though the possibility should not be excluded that it may be possible to find even earlier reports about deadly influenza epidemics either in early Egyptian papyri texts, in texts from Mesopotamia or in texts from China). It most likely happened during the early part of the 23rd century BC, i.e. during the Bronze Age (see Appendix 2 also for a discussion of the date of this incident, the chronological arguments for dating it so early, and the possible identity of the man who was regarded as the founder of Egyptian medicine).]

If this hypothesis is correct, it might help to explain the high virulence of some forms of influenza virus even as far back as during the 1550s [24]. But today, it may be possible that the ecological conditions in farming ecosystems may have changed so much that not even this important virulence character is enough for optimizing the Darwinian fitness of the virus. Two or three completely different high-virulence characters might be even better (helping the virus to win the evolutionary race), which might offer a plausible explanation as to why highly virulent strains of H5N1 virus are even more deadly than the Spanish flu virus.

Reports suggest that there may be at least two additional characters (in addition to the lack of trypsin dependence for haemagglutinin activation) that may contribute to the high virulence of some of the H5N1 strains that are now circulating. The NS gene has mutated so as to enhance the virulence of some of the H5N1 virus strains [32–34]. Pigs infected with recombinant human H1N1 influenza virus that carried the H5N1 NS gene experienced significantly greater and more prolonged viraemia, fever and weight loss than did pigs infected with wild-type human H1N1 influenza virus [33], [34]. These effects required the presence of glutamic acid at position 92 of the NS1 molecule [33], [34].

But it is not entirely clear as yet (in all details) how mutations in the NS gene may affect virulence properties, whether one is dealing with only one molecular mechanism or with a combination of different mechanisms (which appears more likely). If there is more than one mechanism connected with the NS gene that may be important for the high virulence displayed by some of those H5N1 influenza virus strains that are now circulating, this also opens the possibility that there might be more than one type of high-virulence mutation in the NS gene, which might occur either separately (in different H5N1 viral strains) or in combination with each other (in the same viral strain). One of these mutations causes the virus to lose much or all of its normal sensitivity to the antiviral effects of interferons and tumour necrosis factor-alpha (TNF-α) [33], [34]. An important antiviral brake mechanism is thus lost, which could possibly explain why the viral load in human H5N1 influenza patients may rise to a level up to 10 times higher [35] than is found in patients with ordinary influenza.

TNF-α is a pleiotropic signal molecule which has multiple effects; it acts not only on leukocytes but also on various other cell types, e.g. in skeletal muscle [36], [37], the heart [38–41] and adipose tissue [37], [42], [43]. It also acts on endothelial cells [44], [45] and on C-fibers (non-myelinated or very thinly myelinated sensory nerve fibres), which are made more sensitive by the action of TNF-α [46–48]. These effects (and similar effects caused by other cytokines) may help to explain skeletal symptoms and pain in patients suffering from influenza. TNF-α has two different receptors, called TNFR-1 and TNFR-2 [37]. The TNFR-1 receptor has more than one downstream signal transduction pathways. One of the important downstream signal transduction mechanisms is activation of two different sphingomyelinases (with different localization), which are called neutral and acid sphingomyelinase [38], [49–51]. The sphingomyelinases cleave sphingomyelin to ceramide and phosphocholine, and ceramide (N-acylsphingosine) can subsequently be degraded to a fatty acid + sphingosine; these molecules (especially ceramide and sphingosine) may be regarded as functioning as second messengers of TNF-α [40], [50]. Ceramide is important not only as a signal molecule in its own right, but also as a precursor of other molecules also have signal function, such as lactosylceramide [51]. One of the effects of ceramide (and/or of a ganglioside metabolite of ceramide) is to go into the mitochondria, where it interferes with the passage of electrons through the respiratory chain [52], [53]. This causes inhibition of mitochondrial ATP production at the same time as the mitochondrial production of superoxide anion radical is increased (by enhanced leakage of electrons from the respiratory chain through monovalent reduction of molecular oxygen before the electrons are used to make water molecules + ATP in the cytochrome oxidase reaction) [50], [52]. The ratio of reactive oxygen species (ROS) production to ATP production in the mitochondria will thus be enhanced.

When mitochondrial production of ROS is enhanced, these molecules will next function as important signal molecules in their own right; one might possibly call them third messengers. They have multiple signal effects, one of them being activation of the transcription factor NF-kappaB [54–56]. But NF-kappaB has been reported to stimulate replication of the influenza virus and to be necessary for its replication [57], [58]. More replication of the influenza virus (causing enhanced viral load) must in turn be expected to lead to enhanced stimulation of the expression and secretion of various cytokines, including TNF-α. Expression of the TNF-α gene is also partly regulated (and stimulated) by NF-kappaB [59–61]. A vicious cycle will thus be established (or even a double vicious cycle when also taking into consideration the positive feedback loop of ROS- and NF-kappaB-mediated self-activation of TNF-α gene expression), whereby the influenza virus stimulates the production of TNF-α and ROS, whereupon these signal molecules (acting via NF-kappaB) will stimulate the replication of the influenza virus. This vicious cycle can, however, probably be strongly modulated (either weakened or strengthened) by nutrition factors, with good antioxidative nutrient status (e.g. good selenium and GSH status) being protective [62].

When TNF-α normally has two oppositely directed effects on influenza virus replication, i.e. an antiviral effect that may be compared to a brake function and a stimulating effect (via ROS and NF-kappaB) that may be compared to a gas pedal function, it should be easy to understand that it may have dangerous consequences when the brake function is partly or completely lost (because of the mutation in the NS gene). The resulting imbalance will now cause the virus to behave in a way which might be compared to that of a drunk and also very young and inexperienced driver, who is sitting in a car that has lost its brakes. When the car does not stop when the driver tries to push the brake pedal, he tries instead to push the gas pedal even stronger, hoping that this will help him to stop the car – with disastrous consequences.

The theory outlined here, suggesting that the virus, TNF-α, ceramide, ROS and NF-kappaB will be joined together as links in a vicious circle leading to overproduction both of new virus particles, TNF-α and ROS when the antiviral brake function of TNF-α and of interferons is lost, is supported by observations showing that there is not only an overproduction of virus in connection with hypervirulent H5N1 influenza [35], but also an overproduction of TNF-α and other cytokines such as beta-interferon, RANTES (regulated on activation, normal T cell expressed and secreted), interleukin-6 (IL-6) and the chemokine IP-10 [63–65]. A similar overproduction of cytokines (such as IL-1alpha, IL-1beta, IL-6, IFN-gamma and chemokine KC) has also been observed in a mouse model during infection with reassortant influenza viruses that contained NS genes from H5N1/97 and H5N1/01 influenza viruses [66].

With influenza virus strains that respond normally to TNF-α and interferons (so that these signal substances have an antiviral effect), it will presumably not be possible to enhance the stimulation of TNF-α and interferon expression without also enhancing the antiviral effect of these substances, which means they will increasingly act as negative feedback regulators of virus replication. Finally, when their tissue concentration becomes high enough, it may be possible that the braking effect will be so strong that the concentration of viruses in the tissues (or the total number of virus particles per host organism) cannot increase any more. However, this negative feedback effect of TNF-α and interferons on viral replication has been reported to be absent or strongly reduced in NS gene-mutated H5N1 supervirulent strains, which in turn could make it possible for the virus to stimulate the production of much more TNF-α and interferons without becoming inhibited by the antiviral effects of these substances.

Yet another type of high-virulence character that has been found in some of those dangerous H5N1 strains that now circulate in birds is located in viral RNA polymerase complex genes [67], [68]. It was shown in one study that a single amino acid substitution, from glutamic acid to lysine at position 627 of the PB2 protein (which is one of the components of the viral RNA polymerase), was enough to convert a non-lethal H5N1 influenza A virus isolated from a human to a lethal virus in mice [67]. In contrast to the non-lethal virus, which replicates only in respiratory organs, the lethal isolate was found to replicate in a variety of organs in mice, producing systemic infection [67]. When tested in chicken embryo cells and a quail cell line, it was found that the identity of the PB2 amino acid at position 627 did not appreciably affect the replicative efficiency of the virus [67]. However, viruses with lysine at this position instead of glutamic acid grew better in various types of mouse cells that were tested [67]. When the effect of this substitution was investigated in vivo in mice, it was found that all of the test viruses showed the same cell tropism, but infection by viruses containing lysine at position 627 spread more rapidly than those viruses containing glutamic acid at this position [67]. Further analysis showed a difference in local immune responses: neutrophil infiltration in lungs infected with viruses containing lysine at position 627 persisted longer than that associated with viruses lacking a glutamic acid substitution [67]. It was concluded from these observations that the amino acid at position 627 of the PB2 protein determines the efficiency of viral replication in mouse cells (but not in avian cells), but does not affect tropism among cells in different mouse organs [67]. The presence of lysine leads to more aggressive viral replication, overwhelming the host's defense mechanisms and resulting in high mortality rates in mice [67].

Reverse genetics was applied in another study to generate H5N1 reassortants combining genes of lethal A/Vietnam/1203/04 (VN1203), a fatal human case isolate, and non-lethal A/chicken/Vietnam/C58/04 (CH58) and test their pathogenicity in ferrets and mice [68]. The haemagglutinins of the two viral strains were found to have six amino acid differences, identical cleavage sites, and avian-like alpha-(2,3)-linked receptor specificity [68]. Surprisingly, it was found that exchanging haemagglutinin and neuraminidase genes did not alter pathogenicity, but substituting CH58 polymerase genes completely attenuated VN1203 virulence and reduced viral polymerase activity [68]. CH58's NS gene was found to partially attenuate the virulence of VN1203 in ferrets but not in mice [68]. These findings were interpreted by the authors to suggest that to display high virulence in mammalian species, an avian H5N1 virus with a cleavable haemagglutinin may require adaptive changes in polymerase genes to overcome the species barrier [68].

It might be speculated that some mutations in the viral RNA gene polymerase genes will affect the viral RNA synthesis rate (and therefore virulence) only in mammalian cells but not in avian cells, while other mutations may affect the viral RNA synthesis rate in both avian and mammalian cells. The latter type is more likely to be selected for as long as the virus mainly circulates among birds and only rarely infects mammals (including humans).

There may be more than one reason why high-virulence characters could be selected for (enhance the Darwinian fitness of the virus) among those H5N1 strains that now are found to circulate among both domestic and wild bird populations. The first reason is the evolutionary mechanism that has already been mentioned above, i.e. that it happens as a direct consequence of the abnormally high local population density and large total number of individuals per subpopulation under conditions of modern industrial poultry farming, which shifts the position of the fitness optimum when enhanced production of new viruses also may mean an enhanced death risk (or earlier death) for the host.

Another possible reason (or evolutionary mechanism) has to do with the difference between haemagglutinin receptors found in the intestine of ducks and poultry [69], [70]. A Russian group has shown that there is more difference, when comparing the haemagglutinin receptors in the intestine of ducks and of poultry, than between the haemagglutinin receptors of poultry and in the airways of primates [69], [70]. The dangerous strains of H5N1 influenza viruses are now found to infect poultry as well as ducks. This could possibly mean that their haemagglutinins are not perfectly adapted to the receptors found either in ducks or in poultry, but that this may represent some kind of compromise because it is useful for the virus to be able to exploit both species as hosts: poultry because of the large number of poultry flocks and the large size of many of them, which means that each poultry-yard can function as an extremely efficient ‘virus factory’, while wild ducks are important first of all for geographical dispersal of the virus not only over large geographic distances, but also from one farm to another even within the same local district. An analogy might be seen with the different roles played by the adult and larval stages of sessile animals living on the sea-floor, e.g. oysters, where the adult animal is a very prolific ‘factory’ of germ cells, while the non-reproducing larva is important for dispersal.

With a less than perfect adaptation of the haemagglutinin molecule for the receptors of one or both of its alternating host species (which have receptors very different from each other), it might possibly be useful for the virus to compensate for less efficient uptake of virus into the host cells by enhancing the efficiency of viral RNA production, once a virus particle has managed to get inside some target cell. The situation might be compared with a man who is not very attractive or too shy and therefore has been living alone for several years – but once he has finally found a woman who is willing to live with him as his wife, he decides to make up for all those years he has been living without female company by producing as many children as he possibly can. This is a mechanism which possibly could explain changes in the kinetic or regulatory properties of the viral RNA polymerase (provided that this leads to enhancement of the rate of viral RNA production not only in mammalian cells, but also in avian cells), and it may be possible that it also could help to explain the evolutionary emergence of high-virulence properties associated with the NS gene. Reducing the specificity of organ tropism of the virus (i.e. making it possible for the virus to infect a larger number of different organs) might be regarded as an alternative strategy for achieving something of the same (i.e. compensating for poor virus uptake per cell – because of poor species adaptation of the haemagglutinin – by enhancing the total number of potentially available target cells per host individual).

As long as geographic dispersal of the virus depends largely on infection of migratory, wild waterfowl, it may be possible that the selection process will favour haemagglutinins sufficiently adapted for ducks – and therefore maladapted for the alpha-2,6-linked receptor that is found both in the upper and lower airways in humans [71] – that the virus will not be very infective among humans. This could lead to some kind of quasi-stable situation over a certain period of time, when the virus can be very dangerous for those humans that become infected, but not highly contagious (i.e. unlike ordinary influenza, it will not be very easily transmitted to other human hosts).

However, one should not expect that this quasi-stable situation will last indefinitely. There seem to be mechanisms (or possible scenarios) whereby an evolutionary adaptation of the viral haemagglutinin to chicken hosts could take place without much loss of virulence properties, if any. This would involve changing the haemagglutinin so as to make the virus more contagious among chickens, while also facilitating chicken-to-human and human-to-human transmissions. This might, for instance, happen if the chicken population has already experienced an epidemic with some other kind of influenza, offering enough cross-immunity that even if the supervirulent strain could replicate well, it would not be deadly for the birds. It might be possible (though not very probable) that the virus strain arising as an outcome of such a process (with better adaptation of the haemagglutinin to the human-type receptor) could be less virulent than that kind of avian (H5N1) virus that so far has killed more than half of all registered human cases. But it may also be possible, and more likely, that it might become even worse.

Yet another type of evolutionary mechanism that possibly could favour the emergence of more virulent types of influenza virus is also connected with the circulation of completely different types of influenza virus in the wild and domestic bird populations and the possibility of cross-immunization (in some cases), making it possible for other types of influenza virus (other HN numbered combinations) to confer partial immunity against H5N1 influenza. Such cross-immunization can be associated with different types of immune responses, including humoral as well as cellular (T-cell-mediated) immunity. An important mechanism would probably be reaction between the new invaders (H5N1 viruses) and moderately cross-reactive immunoglobulins, which could imply partial but not total blockage of the uptake of H5N1 virus particles into target cells. The consequences of this for viral replication must be expected to be similar to the consequences of less than optimal adaptation of the viral haemagglutinin for the dominating (most common) type of haemagglutinin receptors in a given host species. There might thus be a selective pressure favouring the evolution of compensatory mechanisms, i.e. mechanisms that may help to increase either the total target cell population or the efficiency of viral replication following uptake in the target cells. The evolutionary consequences of this for virulence evolution must be expected to be more or less the same as we have already seen for the case when the haemagglutinin cannot be optimally adapted to the receptors of a given host species (because it is advantageous for the virus to shuttle between two or more different host species that have haemagglutinin receptors that are very different from each other).

It has been shown that other (and less virulent) types of influenza than the highly virulent H5N1 strains, i.e. H9N2 strains, because of immunological cross-reactivity can give partial protection against influenza strains that are highly lethal not only in mammals, but also in birds, so that birds already possessing partial immunity against the lethal strain will not die from the latter [72], [73]. However, the immunity is not strong enough to stop proliferation of the highly virulent strain [72], [73]. Therefore, the mechanism discussed above (i.e. evolution of enhanced virulence as a compensation for cross-immunization) might well be something more than just a theoretical possibility.

It should be noted that this form of cross-immunization also may provide a mechanism whereby influenza virus strains that are extremely virulent for both animals and humans may continue to circulate without killing so many of their animal hosts that an epidemic among birds burns out by itself before it has given rise to a human pandemic. It is theoretically possible that something similar also could happen in mammals, which provides for a mechanism whereby new strains that perhaps could be even more dangerous (because of the combination of very high virulence and ease of transmission among mammalian hosts including humans) could arise and propagate in some mammalian host species (by new reassortment events) without killing off this host too rapidly (so that an epidemic with the new very dangerous ‘super-influenza’ strain among animals may be self-limiting and burn out by itself before humans have become infected).

And it might also provide for a mechanism whereby – theoretically – it would be possible for a supervirulent avian influenza strain coming from ducks or geese to adapt their surface glycoproteins to the receptors in chicken (being similar to the corresponding human receptors) without causing so high a mortality in the chicken population that the epidemic among the chickens either burns out by itself before it has reached very far, or the epidemic is stopped very early because it is discovered by the health authorities in the country concerned who then take drastic measures (mass slaughter and safe destruction of the dead animals) to stop it.

Finally, it should also be noted that the likelihood of flocks of domestic birds (or other domestic animals, e.g. swine) becoming affected by epidemics with different types of influenza (so much that the mechanisms discussed here regarding virulence evolution and the survival of especially virulent influenza virus strains in the bird populations could be of more than theoretical significance) must be expected to depend strongly on various ecological boundary conditions. It may be possible that the importance of these mechanisms may not depend as much on local population density and flock size of the domestic animals as they may depend on the nature and intensity of the contact between domestic and wild bird populations. Good contact between wild and domestic bird populations (e.g. between wild and domestic ducks in several Asian countries) would be expected to favour the introduction of many different influenza virus subtypes (many different HN number combinations) from the wild into the domestic bird populations, which again will enhance the probability of introduction into the domestic bird populations of some of those other HN number combinations that could be capable of providing partial cross-immunity against dangerous forms of H5N1 influenza.

If vaccination were to be carried out against influenza subtypes other than H5N1 in domestic bird populations, this is also a problem that should be very strongly borne in mind. Therefore an influenza vaccine should never be given to birds that may offer partial cross-immunity against some of the highly virulent varieties of influenza virus that are known to circulate among contemporary bird populations, such as those highly virulent strains of H5N1 influenza that are circulating today.

It has sometimes been stated by would-be experts that if and when some highly virulent strain of H5N1 should change so that it becomes highly infective within human populations (through human-to-human transmission), it must be expected that the virus simultaneously will lose much of its virulence properties, so that it most likely will not become more virulent than the Spanish flu virus, and perhaps it will be even milder, e.g. something like the Hong Kong flu pandemic virus. There seems to be no scientific basis for this expectation except an unsupported evolutionary hypothesis that even if it may be qualitatively correct still could be quantitatively wrong. The idea is that it is not advantageous for a pathogen to be too strongly virulent for a given host species. An influenza patient who is no more sick than making it possible for him to go shopping and to travel on the subway or the bus can infect many more other persons than one who is so sick that he will only stay in his bed – or one who is already lying in his coffin.

This type of evolutionary hypothesis is very likely correct, when one considers evolution of the pathogen over a sufficiently long time horizon, e.g. over a time-scale of millions of years, of millenia or of centuries, and maybe also (when we consider the evolution of viruses) over a time-scale of decades. But it is not necessarily correct, not even for rapidly mutating RNA viruses, when we talk about time-scales of years, months, weeks or days.

The problem boils down to comparing the rates of two very different processes. One is the dispersal of a new pandemic virus from the site of initial outbreak to almost all the rest of the world. The other is the process of evolutionary adaptation of the pathogen to a new host species. For both types of processes, it is possible to make mathematical models. But we need to put some numbers into the equations before running a modelling experiment, and the only way to validate the model (i.e. to validate those numbers that we decide to put into it before running the computer experiment) is to be able to compare the outcome of the model calculations with some real-life control (as when doing ‘curve fitting’ for the different waves of the Spanish flu pandemic [1]).

This might be more difficult for the evolutionary processes concerned than it is for the progression of the pandemic itself. However, it should be possible to study in a more direct way (without modelling) those relevant historical examples that might be available. And we have a very good one (but possibly not more than one, which could yield the kind of information we are seeking here) – i.e. the Spanish flu pandemic [74].

The Spanish flu pandemic came in four different waves [74], [75]. Each wave should be considered an independent epidemic, i.e. an independent pandemic. The first wave of Spanish influenza was relatively mild [74–77]. The first cases were reported during the northern spring (March–April) of 1918, but it was not before the end of June that the number of cases skyrocketed [75]. In the first bout of the influenza pandemic during the summer of 1918, the virus spread fast and infected a relatively large majority of the population, but relatively few people actually died from it [75]. The virus apparently mutated during the summer, and at the end of August a second and highly virulent bout of influenza started [75]. Two-thirds of all the deaths from Spanish influenza occurred during the months from October through December 1918 [75]. A third bout of influenza started around the Christmas and New Year celebrations in the year shift 1918–1919 and did not subside before March–April 1919 [75]. The third wave did not spread as fast as the first wave had done and was not as lethal as the second wave, probably because a large proportion of the population had gained immunity and frail individuals had already died in the previous waves [75]. Thus, there were three major outbreaks within the duration of 1 year [75]. But in some places, for example in remote areas in northern Scandinavia and also in some island communities in the South Atlantic, Spanish flu seems to have returned as a fourth wave in 1920 [74], [75], [77]. The mortality rate could locally be especially high during this fourth wave of the pandemic, e.g. 9.8% in Enare in Finnish Lapland [74].

It may be possible that antigenic drift had occurred between different waves of the Spanish flu pandemic, causing enough loss of immune resistance that it was possible for persons who had been infected during some earlier wave to be infected once more in a later wave (e.g. for a person who had been infected during wave 1 to be infected again during wave 3 or wave 4), but with enough partial immunity remaining that there would only be mild symptoms when the same person was infected for the second time. A great majority of such mild cases would probably not have been registered, which means that we cannot expect to find any information about them in the available historical written sources. Maybe there was not much antigenic drift between wave 1 and wave 2 because the time separating wave 1 and wave 2 was too short. It is therefore possible that wave 2 mainly affected persons who had not been infected during wave 1. But the time interval separating wave 1 and wave 3 might have been enough to allow for more significant antigenic drift (being more comparable to the time intervals normally separating successive epidemics of related types of seasonal influenza). It may thus be possible that wave 3 may have affected a much larger fraction of the Earth's population than is apparent from the historical source data, but that most of those who were infected got the disease in a very mild form because of partial immune protection from the first wave. For those who had no immune protection (i.e. had not been infected during either wave 1 or wave 2), it may be possible that the virus was equally as virulent – equally as deadly – during wave 3 as it had been during wave 2. During wave 4, it may be possible that even more antigenic drift had taken place, compared with wave 1. But >90% of the world's population might already have been infected in one of the previous waves when wave 4 started. Most people would therefore get the disease in such a mild form that it would not be noticed. But the high mortality recorded more locally (e.g. in parts of Finnish and Swedish Lapland) during wave 4 tells us that the virus as such was not any less virulent than it had been during wave 2.

What these historical data tell us is that it must have taken at least 2 years from when the Spanish flu pandemic started before an evolutionary adaptation of H1N1 to human hosts causing significant reduction of virulence could have begun. But the Spanish flu pandemic was preceded by outbreaks of respiratory disease in France and the UK in the years 1915–1917 [78]. Certain of these earlier focal outbreaks – called epidemic bronchitis rather than influenza – occurred during the winter months when influenza was known to be in circulation, and presented with a particular heliotrope cyanosis that was also prominent in the clinical diagnosis in the world pandemic outbreak of 1918–1919 (the Great Pandemic) [78]. The outbreaks in army camps at Etaples in France and Aldershot in the UK in 1916–1917 caused very high mortality in 25–35 year olds, and increased deaths from bronchopneumonia and influenza were also recorded in England [78].

It can be deduced from these historical data that early focal outbreaks of influenza-like disease occurred in Europe over a couple of years before 1918 and on the balance of probability the Great Pandemic may have been initiated not in Spain in 1918, but in another European country in the winter of 1916 or 1917 [78], or perhaps even as early as in 1915. It has been proposed that the pandemic had its origins in France, or more precisely on the Western Front, and that conditions there may have facilitated the emergence of this very dangerous form of influenza virus [78–80]. Early outbreaks of a new disease with rapid onset and spreadability that caused high mortality in young soldiers in the British base camp at Etaples in Northern France in the winter of 1917 are, at least to date, the most likely focus of origin of the pandemic [80]. Pathologists working at Etaples and Aldershot barracks later agreed that these early outbreaks in army camps were the same disease as the infection wave of influenza in 1918 [80].

What these historical data do also suggest, however, is that it may possibly have taken as long as 4 or 5 years from the first significant outbreaks of H1N1 influenza in France (in 1915 or 1916?) until human-adapted H1N1 viruses started to evolve in the direction of significantly reduced virulence (after the end of the fourth wave of the Spanish flu pandemic which came in 1920). Maybe it did not happen at all; it may also be possible (at least theoretically) that none of the less virulent H1N1 strains that later continued to circulate as seasonal influenza should be regarded as direct descendants of the Spanish flu virus. Alternatively it might have been another, but far less virulent H1N1 avian-adapted virus strain (i.e. an avian H1N1 virus strain that was dependent on trypsin for replication) that managed to cross the species boundary from birds to humans (most likely by changing host species adaptation of the haemagglutinin first from ducks to poultry, but without acquiring high virulence properties, and next from poultry to humans, perhaps with swine as an intermediate).

It may be concluded that the historical data do not support the hypothesis that H5N1 strains, if or when they evolve so as to become highly infective among humans, will simultaneously lose much of their high-virulence properties. While evolutionary host adaptation of the virus towards reduced virulence might not begin until 2 years or more after the start of the pandemic, model calculations suggest that it might not take more than about 1 month (without vigorous counter-measures) for a new pandemic virus to infect >90% of the world's population and kill half of them [1] – in what might amount to an over-optimistic scenario because it must be judged more probable than not that the real mortality will become even higher.

Why did it take such a long time before H1N1 viruses started to evolve towards significantly reduced virulence (or were replaced by some less virulent grand-cousin) during the period 1915–1920 (or 1916–1920)? It might be speculated that part of the answer could lie in the molecular nature of the high-virulence trait associated with the haemagglutinin precursor molecule. This high-virulence trait was not a result of point mutations leading to one amino acid being replaced by another. It depended on the insertion of some extra amino acids near the cleavage site. If these extra amino acids are to be removed, this cannot happen only as a result of point mutations. A deletion mutation is needed, which also needs to satisfy very stringent specifications (so that all of the extra amino acids are removed, but not more than that, and without causing a frameshift mutation in part of what remains of the haemagglutinin precursor protein gene).

While RNA viruses have high mutation rates, this is presumably mainly in the form of point mutations, while the type of deletion mutation we are talking about here possibly could be much less frequent. On the other hand, the virus may have been successful enough (i.e. may have had good enough infectivity properties – making it possible for it to sweep across the world in three different pandemic waves in no more than about 1 year) that there was no strong reason for it to change. Selective forces may be stronger during difficult times (e.g. in the commercial world during periods of economic depression) than when everything goes exceedingly well.

But the same type of high-virulence trait is also found among those highly virulent H5N1 strains that are now circulating as was found in the Spanish flu virus. The deletion mutation that would be required for removing this high-virulence trait would probably not be easier now than it was during the period 1915–1920. However, H5N1 could be even worse, having genetic properties that must be expected to make it even more difficult (and less probable) for it to change its properties towards significantly reduced virulence once a pandemic has started. While the Spanish flu virus possibly might be compared to a one-headed monster, H5N1 may apparently better be compared to a three-headed one, like the three-headed trolls of some of the Norwegian fairy tales. It may have no less than three different virulence traits connected with three completely different genes (i.e. the haemagglutinin precursor gene, the NS gene and some of the RNA polymerase genes). The probability that all these genes important for virulence will mutate and be selected for simultaneously towards reduced virulence when the virus starts to become more infective among humans (enough to start a new pandemic) must be regarded as extremely low.

H5N1 cannot be expected to start a new pandemic unless the haemagglutinin changes its host species adaptation properties so as to bind better to the human-type (alpha-2,6-linked) receptor. But when this happens, it must more likely than not be expected that it will also make the virus even more virulent than now (because of enhancement of the receptor molecule total number and density per gram of tissue or per organism when not only the alpha-2,3-linked but also the alpha-2,6-linked receptors become available for uptake of the virus into human target cells – which must in turn be expected to enhance the total rate of viral replication per organ or per organism) [1]. If or when this happens, it could therefore function as an additional virulence trait that can be expected to interact synergistically with the others, even if it will not function as a virulence trait alone – when it is not combined with other virulence traits. So the monster will now have four heads instead of three.

When the first serious outbreak of H5N1 influenza among humans was registered in Hong Kong in 1997, there were 6 deaths among 18 registered human cases, i.e. 33% case fatality [81]. During the winter of 2006, the average case fatality (from cumulative numbers for all laboratory-confirmed cases of human H5N1 influenza and all deaths) had increased to 54% – with a total number of 92 human deaths [82]. As of 23 May 2006, there is a cumulative number of 218 laboratory-confirmed human cases and 124 deaths from H5N1 influenza [83], i.e. an average case fatality of 57%. The situation has changed for the better in Vietnam and Thailand, most likely as a result of aggressive counter-measures to get control over the H5N1 epidemic in domestic animals. In Vietnam there were 29 cases and 20 deaths in 2004, 61 cases and 19 deaths in 2005, but no cases so far in 2006 [83]. In Thailand there were 17 cases and 12 deaths in 2004, 5 cases and 2 deaths in 2005 and no cases so far in 2006 [83]. But the situation in Indonesia is not so good. Indonesia had no registered (and laboratory-confirmed) cases in 2004, but 17 cases and 11 deaths in 2005 and 25 cases and 22 deaths in 2006 [83]. The average case fatality in Indonesia has thus increased from 67% in 2005 to 88% in 2006. There has also been at least one significant incident of probable human-to-human transmission in Indonesia, with eight infected and seven dead (but one of these not laboratory-confirmed) within the same family.

The combination of a possible tendency for enhanced human-to-human transmission and enhanced case fatality (from 67% in 2005 to 88% in 2006) that now can be seen in Indonesia must be considered as extremely alarming – as a possible pre-warning that something extremely serious soon might be about to happen – as it may be possible that there could be a common cause for both of them, i.e. an evolutionary process where the haemagglutinin is becoming stepwise better adapted to the human-type (alpha-2,6-linked) receptor than it used to be. But if this should be the case, and if some kind of halfway adaptation of the haemagglutinin for the alpha-2,6-linked receptor is enough to increase the average case fatality to 88%, what should we then expect the case fatality to be when the day comes that the haemagglutinin has become so well adapted to the alpha-2,6-linked receptor that a new pandemic may start? Might we then perhaps be dealing with a case fatality of about 98–99% rather than 88%?

From what is already known about molecular causes of high virulence in this three-headed monster (i.e. those high-virulence strains of H5N1 influenza virus that are now circulating in domestic and wild bird populations), there can be little doubt that we will be dealing with a true superpathogen, that day when the monster may acquire its head number four (which will be needed to make it more infective, but most likely also will make it even more virulent than today). The molecular biological data (about the different molecular causes of high virulence in those strains of H5N1 virus that now circulate) and the high case fatality that has been observed in Indonesia in 2006 seem to tell the same story. The monster is likely to have virulence properties comparable to or worse than the pathogenic agent that was the cause of the Black Death (most likely Yersinia pestis), but it will be even more infective. In the absence of very vigorous counter-measures, it appears more likely than not that the proportion of survivors will be significantly less (possibly by one order of magnitude) than after the Black Death.

Why are governments so little interested in pre-pandemic vaccines?

Immunization against a potential superpathogen would be extremely important, if and when a pandemic superpathogen arrives. Model calculations show, however, that we should not expect it to be possible to use vaccines developed and produced after the pandemic has started, because the pandemic would probably spread so fast all over the world that it would be over before the new vaccine could be delivered, possibly leaving more than half of the world's population dead [1]. A pre-pandemic vaccine might, however, be prepared, assuming that the pandemic virus were antigenically similar to hypervirulent influenza strains now circulating in avian populations [1], [84]. This kind of vaccine could make sense [84] and might possibly help to save a very great number of human lives, provided that large enough stocks of it are available [1] and part of the target population (e.g. health personnel, police, military personnel) has already been vaccinated before the pandemic starts (so that those who should vaccinate the rest of the target population will already be immune themselves at the moment the pandemic starts). Yet, most governments are not enthusiastic about it, and companies are unwilling to make an H5N1 vaccine unless governments pay them to do so [84].

One might ask the question: why should governments not be willing to pay the cost of what actually might be considered an extremely cheap life insurance premium? There seems to be only one plausible explanation for this, i.e. a combination of (i) lack of understanding of the magnitude of the risk (i.e. the product of the magnitude of the potential threat and the likelihood that it could happen), and (ii) a misapprehension of the likelihood that a prefabricated vaccine could work. It may be possible that both the magnitude of the risk and the probability of success for a prefabricated vaccine could be seriously underestimated by most governments, not because of a lack of capacity among politicians to understand this type of question, but because they may have been seriously misinformed by those bureaucratic agencies or scientists who carry the responsibility for giving governments and parliaments the type of information that they need to make correct decisions about such matters.

Possible reasons why the threat has been totally underestimated by governments and parliaments have already been discussed. But what could be the possible explanation for an underestimation of the probability of success for a prefabricated vaccine? It might be speculated that much of the explanation could lie in an inadequate contact between medical research communities on the one side and natural scientists working in environmental and historical disciplines (such as general ecology, evolutionary history and evolutionary biology) on the other: that there might be important misunderstandings regarding the evolution of influenza viruses in bird populations among too many (though far from all!) medical scientists, a consequence of not knowing enough about ornithology, ecology more generally or evolutionary biology (notably in its more quantitative aspects, i.e. regarding the rate at which evolutionary processes normally may proceed). More concretely, it may be possible that there could be two forms of misunderstanding: (i) regarding the rate of antigenic drift for influenza viruses while circulating in animal populations, and (ii) regarding the likelihood that an antigenic shift might occur among avian viruses as a consequence of some form of genetic recombination before a new pandemic starts.

What should be done to improve the world's emergency preparedness for a pandemic of hypervirulent influenza that possibly might come very soon?

The abstract of the article by Moxnes and Christophersen in the previous issue of this journal [1] outlines the principles of a harm reduction strategy which could be the only feasible one (for obtaining significant harm reduction) in a situation where neither enough vaccines nor efficient treatment are available, or when it will not be logistically possible to reach a target population with vaccines within the short time interval that might be available before the superpathogen itself arrives and perhaps will kill nearly everybody. Everything possible must be done to hinder geographic dispersal of the hypervirulent virus (e.g. immediate cessation of all international passenger traffic, immediate cessation of all ordinary road traffic, house quarantine) at the same time as everything possible should also be done to achieve immunization of vulnerable target populations all over the world as soon as possible (so that people may achieve good protective immunity before the superpathogen itself comes and takes them). Vaccines should be used if available (and if they are efficacious enough), but if vaccines are not available or for logistic reasons cannot be used, one may instead – as the only realistic alternative – release a counter-pandemic with a ‘vaccine’ virus that has antigenic properties identical or very similar to the superpathogen, but which is no more virulent than ordinary (seasonal) influenza virus. Isolation/quarantine measures can be gradually lifted after immunization of vulnerable subpopulations has been achieved. Considering as an example the situation in some rural district in a developing country, one can again open for traffic along the roads connecting the villages of this district with each other (and the roads connecting them with the rest of the country) some few weeks after all of the villages concerned have been reached by the counter-pandemic virus, i.e. after a time interval long enough that good protective immunity against the superpathogen will have developed.

However, this is a strategy that requires contingency planning and practical preparations in advance if it is to be possible to do everything that needs to be done swiftly enough and consequently enough – and if it is to be possible to maintain the isolation/quarantine measures for the time that might be needed before immunization can be achieved. It is possible that the only group of professionals that have the experience needed for making that kind of emergency contingency plans that may be called for here – and for swiftly carrying them out when needed – are the military. The best thing that could be done now to deal with a major security threat that possibly could be imminent would therefore be for the Security Council to ask the military forces of all its permanent members to make two types of contingency plans, i.e. (i) some kind of master plan that subsequently could be used as a basis for making national plans by all other countries in the world, and (ii) a more detailed plan for their own territory. The Security Council might possibly also send a similar request to NATO as well as to other regional alliances. It should decide on procedures, how to declare a condition of global emergency, and who should have the mandate to make that decision (which presumably should be made following a recommendation from WHO, but it should be decided in advance whether the Security Council or the Secretary-General should have this mandate – so that there will not be any unnecessary waste of time, which might become very expensive in terms of the number of human lives it easily might cost). When this kind of declaration (of global emergency) comes, it means that orders should be given to the military in all countries in the world to take those actions that will be needed (according to those plans that have been prepared in advance).

People will need food for the time it will take before isolation/quarantine measures can again be lifted. This will be important not only for hindering them from dying from starvation (or from other infectious diseases because of lack of food), but also to minimize morbidity and mortality caused by a ‘vaccine virus’ in a population entirely without immune protection against similar types of influenza (even if the vaccine virus per se is no more virulent than the ordinary seasonal influenza viruses). Having enough of the right types of foods may also be extremely important for improving the chances of survival of those who are not so lucky that they have been immunized before the superpathogen itself arrives. It will therefore be important not only that different countries have enough food to endure a quarantine/isolation period of, say, 3 months, but also that food stores must be sufficiently decentralized to make it possible for local communities (e.g. single villages) and even single households, institutions and companies (e.g. power plants) to endure this quarantine period in total isolation from the rest of the society concerned until immunization can be achieved, until they can receive new food supplies from transport personnel who have themselves been immunized (and have permission to use the roads when showing a certificate of immunization), or until the pandemic is over.

In several poor countries, the present ‘normal’ diet of much of the population cannot be considered adequate for optimal protection against infectious diseases. This could make much of the population in these countries extremely vulnerable if we should experience a new pandemic with something as virulent as the Spanish flu virus or worse. It should therefore be considered a highly urgent task to try to achieve whatever might be possible to improve the food supplies of poor countries as part of the strategy of global emergency preparedness for a possible pandemic with hypervirulent influenza. Emphasis should be not only on improved decentralized stockpiling of foods (so as to make it easier to endure a period of quarantine/isolation), but also enhanced production of food types and specific nutrients that are important for the possibility of surviving malignant forms of influenza, but also in short supply for too many of the world's poor. For many poor countries, it may be especially important to enhance total production of cheap sources of high-quality protein (improving the availability of various essential amino acids now often in short supply in the ordinary diet), as well as improving the availability of a number of micronutrients, including selenium and various B-group vitamins [62].

For household storage in rich countries (making it possible to endure house quarantine for a period of 2 or 3 months), one should possibly consider making dry ‘feed mixtures’ for human consumption in a similar way as is done for domestic animals. These mixtures should be nutritionally fully adequate (contain enough of everything needed), contain no toxic additives (e.g. no harmful synthetic antioxidants) and have an acceptable taste. They should also have very good storage stability, which can best be achieved through package under inert gas (pure nitrogen) in a diffusion-tight bag or container.

The mortality caused by a counter-pandemic virus will probably be higher than during an ordinary influenza epidemic, especially among elderly and malnourished persons [1]. [There must be good reason to hope, however, that it should also be possible to reduce mortality caused by a ‘vaccine virus’ (as well as from ordinary seasonal influenza among elderly and other frail people), using the principles outlined by Christophersen and Haug [62].] Vaccination, using vaccines that have been prefabricated and stored before the start of an eventual pandemic, should therefore be the preferred method of achieving immunization whenever logistically feasible [1]. However, even if various vaccines against such avian-adapted H5N1 influenza virus strains as are now circulating have been tested on humans and found to be safe, they are still not as efficacious as might be desired, and they have not been produced and stockpiled in such quantities as would be needed for effective defence in a pandemic situation. It is therefore extremely important to try to develop new and better anti-H5N1 vaccines as soon as possible and to carry out pre-vaccination (before a pandemic starts) of key personnel (e.g. health personnel, police, military personnel, transport personnel) who will be needed to implement the double strategy of superpathogen containment and rapid immunization if and when a pandemic with some hypervirulent variety of H5N1 influenza should come. No money or effort should be spared in order that this task might be solved as soon as possible.

Conclusion

The world should prepare for the possibility of an unprecedented medical disaster. Homo sapiens may have created a monster, a superpathogen of its own making, with no precedent in recorded human history regarding its combination of infectivity and virulence properties. In a worst case scenario, this superpathogen could be as much as 100% lethal, while being equally as infective as ordinary influenza virus. A more probable scenario could be 98–99% lethality (in the absence of therapy far more efficacious than presently available anti-influenza drugs).

There might be no time to lose. It may be possible that the world has already been living on borrowed time for a long period. Future historians might perhaps consider it a disaster that the Chinese managed to control the H5N1 outbreak in Hong Kong in 1997. If (hypothetically and contrafactually) this outbreak should have got out of control and started a pandemic, it may be possible that consequences would not have been worse than they were during the Black Death, i.e. 60% of the world's population might have died. But the H5N1 virus may have changed to become even more virulent now than it was in 1997, which means that (in the absence of very vigorous counter-measures) the proportion of survivors might perhaps be one order of magnitude less now than after the Black Death: instead of 40% (as after the Black Death), there is a real possibility that only 4% may now survive.

The case fatality among laboratory-confirmed human cases of H5N1 influenza in Indonesia has increased from 67% in 2005 to 88% so far in 2006. At the same time, there are indications also from Indonesia that the virus may be starting to change in a way making it more infective from human to human. We should not disregard the possibility that enhanced infectivity and the enhancement of case fatality from 2005 to 2006 could be two sides of the same coin, i.e. that they could be a consequence of mutations making the haemagglutinin better adapted to the human-type (alpha-2,6-linked) receptor. But the haemagglutinin has apparently still not become adapted well enough that it can start a new pandemic. We should ask ourselves the question: if half-adaptation to the human-type receptor is enough to increase the case fatality rate to 88%, what should we then expect the case fatality rate to be when the haemagglutinin has become fully adapted to the alpha-2,6-linked receptor, so that the pandemic may start?

It may be possible that nothing of what has been described as a putative doomsday scenario above will actually come to happen. But it is the duty of everyone working with emergency contingency planning to try to be prepared for the worst and not just hope for the best. The world should therefore now prepare for a security threat which can only be compared to a full-scale nuclear war (as regards the possible loss of human lives) and that possibly might be imminent.

We wish to thank Professor John G. Crmerod for helpful suggestions during preparation of this article.