key: cord-0042990-rgzin7xi authors: Packer, C.; Altizer, S.; Appel, M.; Brown, E.; Martenson, J.; O'Brien, S. J.; Roelke‐Parker, M.; Hofmann‐Lehmann, R.; Lutz, H. title: Viruses of the Serengeti: patterns of infection and mortality in African lions date: 2001-12-25 journal: J Anim Ecol DOI: 10.1046/j.1365-2656.1999.00360.x sha: 7b0bf82fab65c61cf4cf8530eb2c9fad3b4d2d4c doc_id: 42990 cord_uid: rgzin7xi 1. We present data on the temporal dynamics of six viruses that infect lions (Panthera leo) in the Serengeti National Park and Ngorongoro Crater, Tanzania. These populations have been studied continuously for the past 30 years, and previous research has documented their seroprevalence for feline herpesvirus, feline immunodeficiency virus (FIV), feline calicivirus, feline parvovirus, feline coronavirus and canine distemper virus (CDV). A seventh virus, feline leukaemia virus (FeLV), was absent from these animals. 2. Comprehensive analysis reveals that feline herpesvirus and FIV were consistently prevalent at high levels, indicating that they were endemic in the host populations. Feline calici‐, parvo‐ and coronavirus, and CDV repeatedly showed a pattern of seroprevalence that was indicative of discrete disease epidemics: a brief period of high exposure for each virus was followed by declining seroprevalence. 3. The timing of viral invasion suggests that different epidemic viruses are associated with different minimum threshold densities of susceptible hosts. Furthermore, the proportion of susceptibles that became infected during disease outbreaks was positively correlated with the number of susceptible hosts at the beginning of each outbreak. 4. Examination of the relationship between disease outbreaks and host fitness suggest that these viruses do not affect birth and death rates in lions, with the exception of the 1994 outbreak of canine distemper virus. Although the endemic viruses (FHV and FIV) were too prevalent to measure precise health effects, there was no evidence that FIV infection reduced host longevity. The temporal dynamics of viral pathogens depend on several factors, including transmission properties of the virus, and the survival, immune response and spatial distribution of the hosts (e.g. May 1983; Anderson & May 1991) . In general, viruses that show low pathogenicity and prolonged infectious-ness are likely to exhibit relatively high and constant prevalence (Anderson & May 1979a ) and can therefore be termed`endemic.' In contrast, viruses that spread rapidly, have a shorter duration of infectiousness and cause higher host mortality (or immunity) are more likely to generate discrete epidemics (Anderson & May 1979b) . Epidemic diseases are characterized by rapid changes in the prevalence of infection and may disappear from the host population for extended periods. Such pathogens are likely to require high host densities for invasion and persistence, and the extent of outbreaks should increase with the size of the susceptible host population (May 1983; Onstad & Carruthers 1990; Anderson & May 1982; Jaee et al. 1992) . The population-level impact of viral infections will depend on their eects on host survival and reproduction (May 1983; Dobson & McCallum 1995) . Recent empirical and theoretical studies have shown that virulence can be advantageous to the parasite and is not necessarily a side-eect of novel host±parasite combinations (e.g. Allison 1982; Ewald 1983; Anderson & May 1991; Herre 1993; Lenski & May 1994; Levin 1996) . Although transient impacts of novel parasites can be immense, few studies have addressed the long-term eects of disease in wildlife populations (Grenfell & Gulland 1995) . Most wildlife diseases are studied via incidental post mortems, and manipulative experiments in natural populations are often impossible (Dobson & McCallum 1995) . Viruses have been implicated as the cause of several major declines in carnivore populations (Young 1994) . For example, canine distemper virus (CDV) killed over 70% of the last remaining colony of freeliving black-footed ferrets (Thorne & Williams 1988) and phocine distemper killed over 18 000 harbour seals in northern Europe (Osterhaus & Vedder 1988) . Rabies has been suggested to cause population decline in wolves, Ethiopian wolves and African wild dogs (e.g. MacDonald 1980; Kat et al. 1995; Weiler, Garner & Ritter 1995; Sillero, King & MacDonald 1996; Ballard & Krausman 1997; Laurenson et al. 1998) . Perhaps the best-studied example is the canine distemper virus (CDV) outbreak that struck the Serengeti lions and several other species in late 1993±early 1994 (Roelke-Parker et al. 1996 ; also see Alexander & Appel 1994; Alexander et al. 1995) . CDV originated from the large population of domestic dogs surrounding the Serengeti ecosystem and killed 35% of the Serengeti lions within 6 months. Over 85% of the lions were infected, and many victims showed neurological disorders, encephalitis, and pneumonia. While this outbreak followed a well-de®ned onset, many other viral epidemics have been inferred solely from crosssectional serological surveys (e.g. Olmsted et al. 1992; Spencer & Morkel 1993; Creel et al. 1997 ) and detailed/long-term investigations of virus±carnivore interactions are rare (Grenfell & Gulland 1995; Dobson & McCallum 1995) . In the present study, we evaluate temporal changes in the exposure of African lions to six different viruses. These viruses show a variety of transmission modes, symptoms and durations of infectiousness in domestic carnivores (Table 1) . For example, both feline immunode®ciency virus (FIV, a lentivirus closely associated with HIV) and feline herpesvirus permanently infect their hosts and do not elicit lasting immunity (Table 1 ). In contrast, canine distemper virus (CDV), feline calicivirus, feline coronavirus and feline parvovirus generally cause only temporary infection, although a small proportion of infected hosts may persist as asymptomatic carriers. Using serological data from two different lion populations, we ®rst classify these viruses as either endemic or epidemic. We estimate the timing of each epidemic by combining the temporal pattern of seroprevalence with information on the lions' ages at the time of sampling. We then examine the relationship between the extent of each epidemic, and the proportion and density of susceptible hosts in the population. Finally, we measure the impact of viral infection on lion mortality and birth rates. Lions have been studied continuously in the Serengeti since March 1966 and in the Ngorongoro Crater since 1963 (Packer, Tatar & Collins 1998) . All animals are individually recognized from natural markings (Packer & Pusey 1993 ) and most have been regularly observed since birth (Pusey & Packer 1994) , and thus the precise age is known for each individual. Blood samples were collected in the Serengeti each year between 1984 , and the Crater lions were sampled in 1984 , 1985 and 1991 Table 2 for annual sample sizes). These samples were originally collected for genetic assay (O'Brien et al. 1987; Packer et al. 1991a,b) and to investigate the CDV outbreak in 1994 (Roelke-Parker et al. 1996) . Serological assays were later performed to assess the lions' exposure to FIV (Olmsted et al. 1992; Brown et al. 1994) , CDV (Roelke-Parker et al. 1996) , and feline leukaemia, herpes-, parvo-, corona-and calicivirus . Over 300 lions were also tested for FeLV, but none showed detectable levels of antigens . Table 3 summarizes the precise assays and criteria used to determine sero-status for the remaining six viruses. Many animals were tested repeatedly, and repeat samples are included in any descriptive analysis, but excluded from all statistical tests unless an animal seroconverted. Lions are considered to be`immatures' until 4 years of age. Most females do not start breeding until well after their third birthday, while males do not attain breeding status until they are about 5 years old (Packer et al. 1988 (Packer et al. , 1998 . The Serengeti study area covers the south-eastern quarter of the Serengeti National Park. This 2000 km 2 area consists of two contrasting habitats: acacia woodland and opengrass plains (see Packer et al. 1988) . Lions on the plains live at lower densities and endure lower levels of food availability than their woodlands counter- parts. The Ngorongoro Crater is located about 90 km to the south-east of the Serengeti study area, and the Crater lions enjoy higher and more consistent levels of food availability than the Serengeti lions (Hanby, Bygott & Packer 1995) . However, the Crater is a small, isolated population (80±100 animals) with an associated loss in genetic diversity compared to the much larger Serengeti population (O'Brien et al. 1987; Packer et al. 1991b) , which includes an estimated 2000±3000 animals (Schaller 1972; Hanby & Bygott 1978) . Adult serostatus usually indicates prior exposure to the virus, rather than current infection, since an older animal may still be carrying antibodies from a childhood disease (see below). Therefore, seroprevalence in young animals provides stronger evidence of a recent exposure and, if all the younger cohorts are seronegative, the virus has probably been absent for several years. Under these assumptions, we ®rst use the seroprevalence data to determine whether each virus was`endemic' or`epidemic' over a 10year period. We then provide estimates for the tim-ing of speci®c outbreaks of the epidemic viruses based on sudden changes in the frequency of seropositive hosts. Consistent with a pattern of chronic infection, the lion populations showed constant and high levels of seroprevalence for FIV and feline herpes virus. Rates of FIV infection did not vary signi®cantly across years for either adults or immatures (Fig. 1a ) or by sex (cf. Courchamp et al. 1998) . Combining data from all years, small cubs quickly seroconverted to FIV+ , and this trend was similar in the Serengeti and Ngorongoro Crater (Fig. 2 ). All but two of 374 animals in the Serengeti and Ngorongoro were positive for herpes virus (Fig. 1b) , including all eight cubs sampled before their ®rst birthday. Each of the remaining four viruses appeared to be absent from the Serengeti for 4±12 years followed by a sudden resurgence in the frequency of seropositive hosts (Fig. 3) . The sporadic nature of these widespread infections is most conspicuous in the youngest age classes, and the precise timing of each outbreak diered between the Serengeti and Ngorongoro Crater. CDV. Seroprevalence varied between years in the Serengeti in both adults and immatures (Fig. 3a) . The sharp rise in 1993±94 in CDV seroprevalence of adults and immatures was due to the outbreak documented by Roelke-Parker et al. (1996) , whereas the declining adult seroprevalence in the mid-1980s re¯ected an earlier outbreak. The Ngorongoro Crater lions also appear to have been infected before 1984, with no further epidemics before the end of the sampling period in 1991 (see Fig. 4 ). Parvovirus. Annual seroprevalence of adult lions varied somewhat in the Serengeti, and the seroprevalence of immatures varied signi®cantly among years (Fig. 3b) . Immatures showed the highest frequency of exposure in 1985 (see Hofmann-Lehmann et al. 1996) and in 1992. None of the immatures in Ngorongoro Crater tested positive for parvovirus (see below). Coronavirus. Annual seroprevalence in the Serengeti only varied signi®cantly in immatures (Fig. 3c) . In Ngorongoro Crater, there was evidence of new infections between 1984 and 1991, but our sample size was inadequate to con®rm signi®cant variation between years. Calicivirus. Seroprevalence varied across years in the Serengeti, but varied most dramatically in immatures ( Fig. 3d ). Immatures showed the highest seroprevalence in 1985 and 1990/91. Calicivirus was absent from Ngorongoro Crater . The four`epidemic' viruses showed¯uctuating prevalence in each study population, often increasing in discrete outbreaks. The timing of these outbreaks can be estimated from the seroprevalence data in conjunction with the age of each study animal. For example, out of all the Serengeti lions sampled between 1984 and October 1993, only those animals that had been born before July 1981 were seropositive for CDV (Fig. 4a ). Every lion born after 1981 was seronegative for CDV until 1993/94, when a new epidemic swept through the population, infecting lions of all ages (Fig. 4b ). This pattern strongly suggests that the virus had been absent from the Serengeti from 1981 until the end of 1993. Data from Ngorongoro suggest that CDV infected the Crater lions sometime in 1980, but then disappeared from the population for the following 11 years: the youngest seropositive animal had been born in June 1980, but no animal born later than this date had been infected by the end of our last sampling eort in 1991 (Fig. 4c ). Similar heterogeneities in the age-prevalence data could be discerned for the other three viruses and one example from each virus is plotted in Fig. 5 . We interpret the marked discontinuities in seroprevalence for these viruses to indicate a high rate of infection, similar to the CDV outbreak of 1994 (Fig. 4b) . In support of this assumption, almost all of the putative outbreaks simultaneously infected multiple prides and included cubs as well as adults. The only exception was an outbreak of calicivirus in Crater. Virtually every animal has been infected with FIV by their 4th birthday, although lions in the Serengeti plains show a lower rate of infection than the other two habitats (T = ± 3´54, P < 0´001), even when controlling for age in a multivariate analysis (eect of age: T = 5´94, P < 0´001; eect of Serengeti plains: T = ±3´89, P < 0´001). 1987 that infected only a few members of a single pride located along the northern boundary of our Serengeti study area. The approximate end-dates of outbreaks of each virus are summarized in Table 4 . Except for the 1994 CDV epidemic (where comparable data are not yet available), these dates are presumed to have been the beginning of prolonged intervals where no new seroconversions occurred. Besides these extreme discontinuities in exposure, the seroprevalence data also suggest several earlier outbreaks through a`stair-step' pattern. For example, Fig. 4a shows that virtually all animals born between 1970 and 1977 were seropositive for CDV during the 1984±93 sampling period, whereas animals born in 1978±81 showed a consistently lower prevalence. This pattern may be most easily explained by two epidemics that ended in 1977 and 1981. A similar pattern is evident for calicivirus, where separate epidemics resulted in a higher sero-prevalence for animals born in 1979±86 than for those born in 1987±90 (Fig. 5c ). If these viruses do occur in discrete epidemics, antibody titres and seroprevalence should both decline in the years following an outbreak because the strength of each individual's immune response weakens through time. However, we only resampled enough individuals to test this prediction directly for two viruses. Fifteen parvo+ animals were resampled between the 1985 and 1992 parvovirus outbreaks. Their titre levels declined signi®cantly (n = 13 animals whose titre levels changed, T = 15, P < 0´05 signed-ranks test), and two seropositive animals became seronegative after 1´25 and 4´25 years, respectively. Only two CDV+ animals were resampled in the 1980s, and both showed a decline in antibody titres between 1985 and 1987. Including all animals that were old enough to have been exposed to each virus, seroprevalence consistently declined after the presumed end of each Fig. 3 . Annual seroprevalence in the Serengeti for CDV, parvovirus, coronavirus and calicivirus. Note: the statistical analyses only include those years with sucient sample size (n r 5) to perform chi-square tests. (a) CDV: seroprevalence of adults varied signi®cantly across years (w 2 = 34´65, d.f. = 6, P = 0´0000), but varied even more in younger lions (w 2 = 132´3, d.f. = 5, P = 0´0000). (b) Parvovirus: seroprevalence of adults varied non-signi®cantly across years (w 2 = 11´94, d.f. = 6, P = 0´0633), but varied signi®cantly in younger animals (w 2 = 56´94, d.f. = 5, P < 0´0000). (c) Coronavirus: seroprevalence of adults did not vary across years (w 2 = 1´99, d.f. = 6, P = 0´9203), but varied signi®cantly in younger lions (w 2 = 20´90, d.f. = 5, P = 0´0008). (d) Calicivirus: seroprevalence of adults varied across years (w 2 = 15´66, d.f. = 6, P = 0´0157), but varied more in younger animals (w 2 = 73´62, d.f. = 5, P = 0´0000). epidemic (Fig. 6 ). Responses to three viruses (CDV, calici-and parvovirus) declined at a remarkably similar rate, but coronavirus response declined more rapidly. As predicted, individual titre levels also declined through time for CDV (T = ±8´08, n = 139 seropositive animals, P < 0´0001, r 2 = 0´3209) and parvovirus (T = ±5´96, n = 233 seropositives, P < 0´0001, r 2 = 0´1333), although not for coronavirus (T = ±0´35, n = 179, P = 0´7266). Because of batch eects in the calicivirus assays, ELISA values could not be tested. Note that titre levels also declined with the animal's age at the end of each epidemic for both CDV (P = 0´0162) and parvovirus (P < 0´0001), but the drop through time after each epidemic remained highly signi®cant in a multiple regression model that controlled for age (P < 0´0001, with the eect of age also signi®cant at P < 0´0001). Epidemics should be more frequent and more extensive in large host populations, and there is some indication that viral outbreaks varied according to the overall host population size. Figure 7 shows the size of the study populations in the Serengeti and Ngorongoro Crater over the past 20 + years, as well as the approximate timing of the outbreaks listed in Table 4 . In the Serengeti, both corona-and parvovirus typically occurred in years of high population density. Our study areas include three contrasting habitats with associated dierences in lion density. Calicivirus was absent from the small, isolated Crater population. Prevalence after each coronavirus outbreak was lower (w 2 = 12´63, P < 0´001, n = 221) in the low-density Serengeti plains (where prides are small and pride ranges are 150±500 km 2 ) than in the adjacent woodlands (where prides are large and pride ranges are 40±75 km 2 ). Age-prevalence of FIV was also lowest in the plains (Fig. 2) ; a pattern similar to domestic cats where FIV prevalence is lower in low-density areas (Malik et al. 1997) . Although total host density was associated with changes in disease prevalence in the above cases, epidemiological theory speci®es that only the suscep-tible subset of hosts will regulate the invasion and rate of spread of viral infection. In the following sections, we therefore estimated the size of the susceptible population for each virus through time, and we use these estimates to evaluate the association between susceptible host density and viral outbreaks. Because our serological survey included a small subset of the study populations and our sample size varied from year to year (Table 2) , we can only provide rough estimates of the overall number of susceptibles. We base these estimates on the following simplifying assumptions: (i) every animal in the population was exposed to the virus during each outbreak; (ii) survivors of each outbreak gained immunity that persisted over their remaining life-spans; thus (iii) only animals born after the most recent outbreak were`susceptible' to viral infection. These assumptions clearly underestimate the size of the susceptible population because, ®rst, it is unlikely that 100% of the population is exposed to the virus during an outbreak, and, secondly, a proportion of seropositive animals may lose their immunity over time (as implied by Fig. 6) . However, in the absence of complete information, this procedure at least provides a standardized estimate of the minimum number of susceptibles in the population. The total number of`new susceptibles' (i.e. individuals born after an epidemic who had not yet been exposed to the virus) is presented in Fig. 8 . Note that this analysis is restricted to the Serengeti, since estimates require at least two outbreaks of the same virus. For parvovirus and calicivirus, successive outbreaks occurred shortly after the population crossed an apparent threshold number of new susceptibles: parvovirus outbreaks occurred after the number of new susceptibles reached 150 animals, whereas calicivirus outbreaks required at least 90 new susceptibles. In contrast, there were no consistent patterns for CDV (with the 1981 outbreak coinciding with only 70 new susceptibles vs. 250 susceptibles in late 1993) or coronavirus (approximately 80 susceptibles in 1988 vs. approximately 160 in 1992). In the case of CDV, it is important to note that the lions should not be measured in isolation: CDV infected large numbers of Serengeti canids in the late 1970s and early 1980s, so the true number of susceptibles in the ecosystem should also include these other host species. Table 4 . Dates when viral epidemics are presumed to have ended.`Pattern' distinguishes between two situations;`presence/ absence' indicates that consecutive cohorts were seronegative whereas older cohorts were largely seropositive; and`stairstep' designates when consecutive cohorts were moderately seropositive, while older cohorts showed even higher seroprevalence.`Signi®cance' refers to the dierence in seroprevalence between cohorts born within the 4 years prior to the endpoint vs. those born in the ®rst 4 years thereafter (as tested by w 2 with 1 d.f.). The 1976 parvovirus, and 1988 and 1993 coronavirus epidemics all showed the same qualitative pattern as the statistically robust outbreaks, and are therefore included in subsequent analyses. However, the 1987 calicivirus outbreak is omitted from further consideration, because only a small proportion of at-risk individuals became seropositive, and all three of these animals lived in the same pride at the edge of the study area Presence/absence P < 0´001, n = 54 Fig. 6 . Seroprevalence plotted as a function of the number of years that had elapsed after the presumed end of each epidemic. Data are restricted to animals who were alive during the putative outbreak and thus who could have been exposed to the virus. The decline in seroprevalence through time was statistically signi®cant for all four viruses (CDV: T = ±4´88, P < 0´001, n = 223, logistic regression; parvovirus: T = ±4´81, P < 0´001, n = 277; coronavirus: T = ±2´46, 0 = 0´014, n = 293; calicivirus: T = ±4´31, P = 0´001, n = 240). These all remain signi®cant when other age eects are included in a multivariate logisticregression analysis. To estimate the proportion of susceptibles that became infected during each outbreak, we recalculated the annual seroprevalence of the Serengti lions. In this analysis, we assumed that seronegative animals had never been infected between birth and the time they were sampled. We assumed that seropositive animals had been seropositive each year following the most recent outbreak for that particular virus, but we made no assumptions about their serostatus prior to the most recent outbreak. In this way, the age and serostatus of sampled lions was used not only to estimate seroprevalence in the year sampled, but also in previous years (thus reducing the random variance produced by small sample sizes). These estimates are presented in Fig. 9 , and they are broadly similar to the raw data presented in Fig. 3 . However, these estimates interpolate data for the years 1988±92 (when few samples were collected) and extrapolate backwards to years before our serological samples were collected. Most important, Fig. 9 provides an approximation of the magnitude of each outbreak. For example, the 1993±94 CDV outbreak struck when the population was entirely seronegative and over 85% of the population seroconverted in the course of the epidemic. In contrast, the 1990 calicivirus epidemic struck when nearly 30% of the population was seropositive for calicivirus, and seroprevalence only increased to about 50%. We used these data to estimate the extent of each outbreak. We de®ne`incidence' as the proportion of susceptibles that became seropositive during the time course of a single outbreak. Thus, an increase in seroprevalence from 20 to 90% indicates a reduction in susceptibles from 80 to 10%, and an`incidence' of 0´70/0´80 = 0´875; an increase in seroprevalence from 55 to 60% would give an`incidence' of 0´05/0´45 = 0´111. Using data from all four viruses, Fig. 10 shows that epidemics with the highest`incidence' were associated with the highest initial proportion of susceptibles. This trend verges on statistical signi®cance if restricted to the well-de®ned epidemics listed in Table 4 (arcsine-transformed linear regression: T = ±2´49, r 2 = 0´5535, n = 7, P = 0´0552), and the relationship becomes highly signi®cant if we include the 1987 calicivirus Table 4 ) and horizontal lines show the span over which samples were collected. `outbreak' that did not spread beyond a single pride (T = ±4´21, r 2 = 0´7434, n = 8, P = 0´0056). Although the overall trend in Fig. 10 is quite striking, the data are based on very rough approximations and on only 1±3 outbreaks per virus. Within each virus, only calicivirus shows a trend in the same direction as the overall pattern. The paired outbreaks of parvovirus and coronavirus show slight trends in the opposite direction to the overall pattern, but it is perhaps more noteworthy that the two points in each pair are so similar. This suggests that epidemics of parvovirus and coronavirus each spread with a typical magnitude and each strike at a typical level of seroprevalence. Noteworthy, too, is the possibility that parvovirus may consistently infect a higher proportion of susceptibles than does coronavirus, perhaps due to a longer persistence outside the host (weeks for parvovirus vs. hours for coronavirus). We can only provide limited data on the precise cause of mortality in these animals. Most lions sim-ply disappear from the study area, and most observed fatalities result from direct combat with other lions or from injuries sustained during prey capture (Packer et al. 1988 ). The only clear-cut examples of death due to disease occurred during the CDV outbreak in 1994 (Roelke-Parker et al. 1996) . In the following section, we rely on a demographic analysis to determine whether any of these viruses is associated with widespread mortality or infertility. These assessments depend on whether the virus is endemic or epidemic. Because only two lions tested negative for herpesvirus, we lack a control group with which to compare infected animals. We are therefore unable to measure eects of herpes infection on host ®tness. Although virtually every lion was FIV+ by the age of 4 years (Fig. 2) , about 20% of 1±2-year-olds were FIV±; thus, we restrict our analysis to the long-term survival of young animals. In domestic cats, FIV infection is permanent and in¯icts most mortality 3±5 years after the initial infection (see Fig. 8 . Estimated number of`new' susceptibles for each virus each year in the Serengeti.`New susceptibles' only include animals that were born subsequent to an outbreak and they only remain susceptible until the next successive outbreak, whereupon further cubs contribute to the next set of new susceptibles. Table 1 ). If FIV is similarly harmful to lions, we would expect FIV+ lions to suer a shorter life expectancy than FIV± lions of a comparable age. Note, though, that all the FIV± animals presumably contracted the virus by their fourth birthday, and our analysis therefore contrasts the survival of animals infected early in life vs. those that would have been infected at a later age. Figure 11 shows the annual age-speci®c mortality of lions according to their FIV status as yearlings (top) or as 2-year-olds (bottom). Data are plotted separately for males and females because of their contrasting life histories as subadults (Packer et al. 1988 ). FIV did not measurably increase mortality either in the short term (when FIV± animals were presumably still uninfected) or in the long term (overall, seven FIV+ animals survived > 10 years). Although these data are too limited to detect subtle eects on mortality, we have seen no clinical signs of immune de®ciency except during the CDV outbreak in 1994. CDV is immunosuppressive, but there was no association between clinical pathology and FIV status of CDV victims (Roelke-Parker et al. 1996) . Further, we found no evidence that co-infection with any of the other viruses in¯uenced longevity of FIV+ individuals (all statistical tests non-signi®cant). The entire adult population was FIV+ , so we cannot assess the eects of viral infection on fecundity. Fig. 9 . Estimated annual seroprevalence for each virus in the Serengeti. All seronegatives are assumed to have been seronegative for their entire lives. Seropositives are assumed to have been exposed to the virus at the most recent outbreak, but no assumptions are made about their earlier serostatus. Fig. 10 . The`incidence' of each Serengeti outbreak plotted as a function of the estimated seroprevalence at the onset of the epidemic. Incidence is de®ned in the text. The 1987 calicivirus`outbreak' is plotted in parentheses, since it did not spread beyond a single pride. The sporadic appearance of each epidemic virus enabled us to test whether epidemics coincided with overall reductions in lion mortality or fecundity. For this analysis, we only examine the survival and reproduction of individuals that could have been susceptible to the virus at the onset of each epidemic. Outbreaks of each virus occurred at 4±12-year intervals (Table 4) . We therefore restricted our analysis to the annual mortality of lions <4 years old, since these cohorts consisted entirely of`susceptibles' at the onset of each outbreak. Cub mortality is of particular interest here, since most disease-induced mortality is restricted to kittens in domestic cats (Table 1) . Survival rates in young lions depended on age and¯uctuated from year to year (Fig. 12) , but there was no consistent pattern associated with any particular virus. For example, cub survival was sig-ni®cantly reduced during the calicivirus outbreak of 1980 (w 2 = 19´35, d.f. = 1, P = 0´0000). Although cub survival was also below-average in the calicivirus outbreak of 1985, cubs enjoyed higher survival than average during the 1990 calicivirus epidemic. Only the 1994 CDV outbreak in¯icted lower survival on every age-group (with signi®cant declines for yearlings (w 2 = 24´09, d.f. = 1, P = 0´0000) and 3year-olds (w 2 = 10´61, d.f. = 1, P = 0´0011). However, neither the 1981 CDV outbreak in the Serengeti nor the 1980 outbreak in Ngorongoro had any appreciable impact on lion survival, although two Serengeti lions developed myoclonus in 1981. Without clear signs of illness, we cannot rule out the possibility of a spurious association between a single disease outbreak and a year of low survival. However, the CDV outbreak of 1994 was well-documented and included clearly observed cases of seizures and myoclonus (Roelke-Parker et al. 1996) . This suggests that the high survival rates during earlier outbreaks of CDV may have been due to nonpathogenic strains of the virus. It was not possible to estimate the extent of the earlier CDV epidemics, but it is at least noteworthy that CDV in 1994 was the most extensive of any documented outbreak and struck the largest population of susceptibles. Fig. 11 . Age-speci®c mortality rates of (a) females and (b) males according to their FIV status when sampled between the ages of 1±2 years (top) and 2±3 years (bottom). Sample sizes of FIV+ animals are given in normal type, of FIV± animals in italics. There was no signi®cant eect of FIV status on the mortality rates of either sex or either age-class or in the combined data set. Unfortunately, we have been unable to collect any blood samples from Ngorongoro Crater since 1991, so we cannot test whether the recent decline in this population (see Fig. 7 ) was associated with a viral outbreak. A domestic dog with CDV was found on the Crater rim in October of 1994 (Cleaveland 1996) , and several of the Ngorongoro Crater lions were conspicuously ill at about the same time, but none showed neurological symptoms of CDV. We measured`fecundity' as the proportion of females that gave birth during a particular year. Because of high adult seroprevalence for all viruses, adult females could only have been`susceptible' to a virus if the interval between outbreaks was suciently long. This analysis is therefore restricted to seven widely-spaced outbreaks in the Serengeti and only includes young (and hence susceptible) females. Because of age-related changes in fecundity (Packer et al. 1998) , as well as medium-term ecological variations , we compare the reproductive performance of 4±13-year-old females over the ®rst 2 years before and/or after each outbreak. As in the mortality analysis, none of the four viruses consistently in¯icted a signi®cant loss in fecundity, although there seems to be a recurring eect of CDV (Fig. 13 ). Again, it is possible that there were strain dierences, as fecundity was slightly lowered during the 1985 parvovirus and calicivirus out-breaks. The high annual variation around the 1990 calicivirus and 1993 coronavirus epidemics is dicult to interpret and presumably re¯ects other ecological factors (such as annual variations in prey availability, population density, and social disruption; see Packer et al. 1988) Temporal variation in seroprevalence indicates that both FIV and herpesvirus are endemic in these lion populations. Both show consistently high prevalence (Figs 1 and 2) and both appear to be relatively harmless, although it is dicult to compare the ®tness of infected and uninfected hosts. In domestic cats, FIV follows a similar aetiology as HIV in humans (Pedersen et al. 1987; Yamamoto et al. 1988; Torten et al. 1991 ) and, like HIV in humans, FIV appears to have infected domestic cats only recently (Rigby et al. 1993; Carpenter & O'Brien 1995) . We could ®nd no evidence that FIV status in¯uenced lion mortality (Fig. 11) , nor did we observe any obvious signs of immunode®ciency or clinical pathology in FIV+ individuals (but see Poli et al. 1995) . Phylogenetic analysis of FIV sequences in our study animals suggests that lions have been hosts to the virus for long periods of time (Brown et al. 1994) , and an attenuated eect from FIV infection may have arisen through a variety of co-evolutionary mechanisms (reviewed by Carpenter & O'Brien 1995) . Alternatively, studies of domestic cats suggest that FIV may be more pathogenic in individuals that are co-infected with FeLV (Courchamp et al. 1997) . Thus, the low impact of FIV in these lions may be because they are not co-infected with FeLV. The remaining four viruses showed high temporal variation in seroprevalence, indicative of epidemic outbreaks of disease (Figs 3 and 4) . These outbreaks typically occurred after a large number of susceptibles were recruited into the population (Fig. 8) , and the extent of each outbreak apparently depended on the size of the susceptible host population (Figs 9 and 10 ). Previous studies have documented the existence of threshold host densities for the establishment of pathogens in natural populations (Jaee et al. 1992; Dobson & Hudson 1995; Dobson & Meagher 1996) . In the lions, calicivirus, parvovirus, and to some extent coronavirus showed evidence that viral invasion required a threshold density of susceptible hosts (Fig. 8) . Minimum estimates of susceptible host density varied among the viruses, and ranged from 75 to 200 animals in the Serengeti study area (which includes about 10% of the overall Serengeti lion population). Viruses that were associated with higher initial rates of spread (as estimated by our measure of incidence') also required the highest threshold densities of seronegative hosts. For example, calicivirus and coronavirus outbreaks occurred when susceptible host densities were relatively low and were associated with smaller increases in seroprevalence. In contrast, parvovirus required a larger number of susceptibles and infected a larger proportion of susceptibles during outbreaks. CDV is both the exception and the rule for this pattern. CDV infected numerous carnivore species in the late 1970s, as well as in 1994 (reviewed in Roelke-Parker et al. 1996; Cleaveland 1996) , so the`susceptible population' of only 75 lions in 1981 and over 200 lions in 1994 actually included uncounted numbers of susceptible canids, hyenas and other large felids. However, the lion population was not only large, but also 100% susceptible during the 1994 CDV epidemic, and over 85% of the lions were infected within a few months. None of the epidemic viruses consistently lowered host ®tness. The 1994 CDV outbreak was conspicuously harmful (as emphasized by the sharp decline Fig. 13 . Annual female fecundity before, during and after associated epidemics. Numbers indicate sample sizes and arrows indicate the years in which the epidemics occurred. For years prior to each outbreak, data only include adult females that had not yet been exposed to the virus. In the year of an outbreak, the data only include adult females who were exposed to the virus for the ®rst time that year. For the following years, data include all adult females who had been exposed for the ®rst time during that outbreak (including females who had been exposed as subadults). Analysis only includes data-points based on at least ®ve females. Annual fecundity varied signi®cantly in the years around the 1993 coronavirus outbreak (w 2 = 14´37, d.f. = 3, P = 0´0062) and the 1990 calicivirus outbreak (w 2 = 8´89, d.f. = 3, P = 0´0308). in the Serengeti population, see Fig. 7 ), and the 1980 calicivirus epidemic may have in¯icted considerable cub mortality, but other outbreaks of these same viruses appeared to be essentially harmless. Indeed, none of these viruses regularly in¯icted the same degree of mortality observed in domestic cats (Table 1) . Sequence data are only available for the 1994 CDV strain (Roelke-Parker et al. 1996) , so we cannot make similar assessments of co-evolutionary history as for FIV, nor can we comment on genetic dierences between strains of the same virus. However, strain dierences in pathogenicity have been well documented in other morbiliviruses besides CDV (e.g. rinderpest, Plowright 1982) . It is also possible that all of these viruses were harmful to some extent, but our demographic analysis was too coarse-grained to detect their eects. The timing of each outbreak was estimated retrospectively, we could not control for confounding interactions with other ecological variables, and the lions were no doubt co-infected with additional pathogens and parasites. No ®eld analysis could ever provide the sort of comparisons that would be available from controlled experiments (Dobson & McCallum 1995) . Dobson & McCallum (1995) and Jaenike (1998) have shown that host populations can be limited by pathogens that in¯ict moderate eects on host mortality and moderate to large eects on fertility. Out of all the viruses included in this study, only CDV appears to be suciently virulent to cause measurable declines in the Serengeti/Ngorongoro lion populations. CDV caused considerable mortality in 1994 and may have lowered host fertility during the 1981 and 1994 outbreaks. However, CDV entered the lion population in discrete, widely-spaced epidemics, and these epidemics have been suciently infrequent to in¯ict lasting consequences. The Serengeti population recovered rapidly from the 1994 epidemic, regaining its former size by the middle of 1997 (Fig. 7) . Domestic dogs are the primary reservoir for CDV in the Serengeti/Ngorongoro region (Cleaveland 1996) , and the dog population has grown dramatically with the growing human population over the past 30 years, raising the prospects of more frequent, extensive, and virulent CDV epidemics in the lions and other wild carnivores. Attempts to inoculate this reservoir against CDV are currently underway (Kaare & Cleaveland 1997) . African wild dogs (Lycaon pictus) endangered by a canine-distemper epizootic among domestic dogs near the Masai-Mara National Reserve Evidence of canine distemper virus infection among free-ranging spotted hyenas (Crocuta crocuta) in the Masai Mara Coevolution between hosts and infectious disease agents, and its eects on virulence Population biology oectious diseases. Part I Population biology oectious diseases. Part II Directly transmitted infectious diseases: control by vaccination Infectious Diseases of Humans Occurrence of rabies in wolves of Alaska A lion lentivirus related to feline immunode®ciency virus: epidemiologic and phylogenetic aspects Coadaptation and immunode®ciency virus: lessons from the Felidae The epidemiology of rabies and canine distemper in the Serengeti Dynamics of two feline retroviruses (FIV and FeLV) within one population of cats At-risk individuals in feline immunode®ciency virus epidemiology: evidence from a multivariate approach in a natural population of domestic cats (Felis catus) Serosurvey for selected viral diseases and demography of African wild dogs in Tanzania Microparasites: observed patterns in wild animal populations. Ecology of Infectious Diseases in Natural Populations Detecting disease and parasite threats to endangered species and ecosystems The population dynamics of brucellosis in the Yellowstone National Park Host-parasite relations, vectors, and the evolution of disease severity Introduction: Ecological impact of parasitism on wildlife host populations Population changes in lions and other predators Ecology, demography and behavior of lions in two contrasting habitats: Ngorongoro Crater and the Serengeti Plains. Serengeti II: Research, Management and Conservation of an Ecosystem Population structure and the evolution of virulence in nematode parasites of ®g wasps Prevalence of antibodies to feline parvovirus, calicivirus, herpesvirus, coronavirus and feline immunode®ciency virus in sera from lions in East Africa On the capacity of macroparasites to control insect populations Density-dependent host±pathogen dynamics in soil microcosms Targeting rabies control in the Serengeti: the rationale and design of a domestic dog vaccination campaign Rabies and African wild dogs in Kenya Disease as a threat to endangered species: Ethiopian wolves, domestic dogs and canine pathogens The evolution of virulence in parasites and pathogens: reconciliation between two competing hypotheses The evolution and maintenance of virulence in microparasites Rabies and Wildlife Prevalence of feline leukemia virus and feline immunode®ciency virus infections in cats in Sydney Parasitic infections as regulators of animal populations Biochemical genetic variation in geographically isolated populations of African and Asiatic lions Worldwide prevalence of lentivirus infection of wild felidae species: epidemiologic and genetic aspects Epizootiological models of insect diseases Identi®cation of virus causing recent seal deaths Reproductive success of lions Case study of a population bottleneck: lions of Ngorongoro Crater A molecular genetic analysis of kinship and cooperation in African lions Reproductive cessation in female mammals Isolation of a T-lymphocyte virus from domestic cats with an immunode®ciency±like syndrome The eects of rinderpest and rinderpest control on wildlife in Africa Lentivirus infection in an African lion: a clinical, pathologic and virologic study Non-ospring nursing in social carnivores: minimizing the costs Evolution of structural patterns of feline immunode®ciency virus: molecular epidemiology and evidence of selection for change A canine distemper virus epidemic in Serengeti lions The Serengeti Lion Rabies and mortality in Ethiopian wolves (Canis simensis) Serological survey of sera from lions in Etosha National Park Disease and endangered species: the black-footed ferret as a recent example Progressive immune function in cats experimentally infected with feline immunode®ciency virus Occurrence of rabies in a wolf population in northeastern Epidemiologic and clinical aspects of feline immunode®ciency virus infection in cats from the continental United States and Canada and possible mode of transmission Pathogenesis of experimentally induced feline immunode®ciency virus infection in cats Natural die-os of large mammals: implications for conservation Data analysis was supported by the NSF program in Long-Term Research in Environmental Biology (NSF grant DEB-9407880). Serological surveys were supported by NSF (BSR-8507987, DBS-8807702 and IBN-9107397), the Smithsonian NOAHS Center, Messerli Foundation, Cornell University, and the Union Bank of Switzerland. We thank the anonymous referees for their extensive comments.