Cowpox virus infection in natural field vole Microtus agrestis populations: delayed density dependence and individual risk


and present address: Michael Begon, School of Biological Sciences, The University of Liverpool, Biosciences Building, Crown Street, Liverpool, L69 7ZB, UK. E-mail:


  • 1Little is known about the dynamics of pathogen (microparasite) infection in wildlife populations, and less still about sources of variation in the risk of infection. Here we present the first detailed analysis of such variation.
  • 2Cowpox virus is an endemic sublethal pathogen circulating in populations of wild rodents. Cowpox prevalence was monitored longitudinally for 2 years, in populations of field voles exhibiting multiannual cycles of density in Kielder Forest, UK.
  • 3The probability that available susceptible animals seroconverted in a given trap session was significantly positively related to host density with a 3-month time lag.
  • 4Males were significantly more likely to seroconvert than females.
  • 5Despite most infection being found in young animals (because transmission rates were generally high) mature individuals were more likely to seroconvert than immature ones, suggesting that behavioural or physiological changes associated with maturity contribute to variation in infection risk.
  • 6Hence, these analyses confirm that there is a delayed numerical response of cowpox infection to vole density, supporting the hypothesis that endemic pathogens may play some part in shaping vole cycles.


Host populations exhibiting multiannual cycles of density are good systems for investigating the impact of endemic pathogens on host dynamics, due to the clear dominance of pattern over noise. They also allow parasite dynamics to be investigated over a wide and predictable range of host densities, promoting the detection of any density dependence in either transmission or prevalence. Host density is likely to influence pathogen dynamics, because the probability of contacts between susceptible and infectious individuals may increase with host density (De Jong, Diekmann & Hesterbeek 1995) and/or individuals may exhibit changes in behaviour that alter their susceptibility to infection, including increases in aggressive encounters or in movement. This study examines the dynamics of endemic cowpox virus in cyclic populations of one of its reservoir hosts, the field vole Microtus agrestis L., in Kielder Forest, northern England. Pathogens have been suggested as potentially having an important role in affecting cyclic host dynamics (Mihok, Turner & Iverson 1985; Soveri et al. 2000; Cavanagh et al. 2004), and an alternative potential cause of field vole density cycles at Kielder, the specialist predation hypothesis, has been rejected experimentally (Graham & Lambin 2002). Also, throughout the multiannual cycle, voles exhibit changes in body mass, and age and mass at maturity (Krebs & Myers 1974), yet in translocation studies at Kielder, the life histories of translocated voles shift to those of residents. This argues against genetic differences or maternal effects and in favour of environmental influence mediated by either resources or natural enemies (including pathogens) (Ergon et al. 2001b; Ergon, Lambin & Stensteth 2001a). The possible role of pathogens in the cyclic dynamics of field voles at Kielder warrants further study.

Cowpox virus is an orthopox-virus endemic in rodent populations throughout Europe and western Asia (Baxby & Bennett 1999). In the UK, the highest seroprevalence occurs in field voles, bank voles Clethrionomys glareolus (Schreber) and wood mice Apodemus sylvaticus L. and these species are viewed as the reservoir hosts (Chantrey et al. 1999). The mechanism of cowpox virus transmission has not been demonstrated experimentally, but a variety of evidence suggests that the virus is transmitted directly to susceptible hosts from infectious individuals (Baxby & Bennett 1999; Robinson & Kerr 2001; Carslake et al. 2005). Crucially, previous studies on noncyclic populations of bank voles and wood mice have demonstrated that the virus affects fecundity by delaying the onset of reproduction in both the laboratory and field (Feore et al. 1997; Telfer et al. 2005) and also increases the probability of host survival in the wild in summer, and decreases it in winter (Telfer et al. 2002), possibly as a result of infected individuals delaying reproduction and therefore avoiding associated energetic and predation costs. Cavanagh et al. (2004) suggested a pattern of delayed-density dependence, with a lag of 3–6 months, between field vole density at Kielder and the prevalence of cowpox virus infection, but used only serological data from a limited number of independent samples. Hence, current infection levels could not be distinguished clearly from past infection, making it difficult to determine the processes underlying the observed patterns.

The present study investigates cowpox prevalence longitudinally, examining samples taken repeatedly from marked individuals, enabling current infections to be separated from past infections. It aims to assess whether infection prevalence was density-dependent, whether this was direct or delayed, and the length of any time-lag. This is important for evaluating a pathogen's potential impact on the host population. Direct density dependence stabilizes population dynamics, whereas delayed density dependence, in conjunction with stochastic perturbations and moderate levels of direct density dependence, may cause multiannual cycles of density (Royama 1992; Bjørnstad, Falck & Stensteth 1995). This study investigates specifically the relationship between host density and the per capita risk of infection with cowpox virus, and also how covariates, such as the sex, age or reproductive maturity affect the risk of an individual acquiring cowpox infection, as these too are crucial in determining the detailed effects of the pathogen on the host at both the individual and population level. This is the first time that sufficient longitudinal data have been available to investigate individual covariates and infection status for any endemic pathogen in a natural wildlife population.

Materials and methods

Work took place in Kielder Forest, a man-made spruce forest occupying 620 km2 on the English–Scottish border (55°13′ N, 2°33′ W). Field voles inhabit grassy clear-cuts (16–17% of the total area) but are completely absent from forested areas. Clear-cuts range in size from 5 to 100 ha, with habitat dominated by Deschampsia caespitosa Beauv., Agrostis tenuis Sibth., and Juncus effusus L. Field vole populations at Kielder fluctuate cyclically with a 3–4-year period (Petty 1992; Lambin, Petty & McKinnon 2000). Populations situated close together fluctuate in synchrony, but at a wider scale populations are out of synchrony (Lambin et al. 1998). Voles were trapped in four similar-sized clear-cuts, in two areas of the forest approximately 12 km apart, between May 2001 and July 2003. KCS and PLJ were situated 4 km apart, with vole populations at low to increasing density during the study. BHP and ROB were 3·5 km apart, with voles at increasing and peak density.

Populations were trapped (‘primary sessions’) every 28 days (apart from winter, November to March, when primary sessions were 56 days apart). Each site had a permanent 0·3 ha live-trapping grid consisting of 100 Ugglan Special Mousetraps (Grahnab, Marieholm, Sweden), set at 5 m intervals and baited with wheat grain and carrot. In each primary session, traps were pre-baited for 3 days, set at approximately 18.00 h on the first day and checked five times (‘secondary sessions’) at roughly 12 h intervals at dawn and dusk.

Individuals were identified using subcutaneous microchip transponders (AVID plc, Uckfield, East Sussex, UK) injected under the skin at the back of the neck. Mass, sex and reproductive status were recorded at the time of first capture in each primary session. Animals with juvenile fur or in their first moult were classed as juveniles (Graham & Lambin 2002). A multiple-regression analysis was used to identify a mass threshold for assigning animals with adult coats to juvenile or adult categories, using monthly growth rates estimated from field data, in conjunction with laboratory information on mass at 2 weeks of age (Burthe 2005). Animals were assigned to reproductive classes according to the external appearance of reproductive organs. A body condition score was calculated for each animal by estimating the degree of fat cover over the vertebral column and dorsal pelvic bones. Each area was scored on a scale between 1 and 5 and scores summed (Cavanagh 2001).

A 20–30 µL blood sample was taken from the tail tip of each individual each primary session. Antibody to cowpox virus was detected in sera by immunofluorescence assay (Crouch et al. 1995). Briefly, sera were diluted 1 : 10 and 1 : 20, and incubated on fixed, cowpox virus-infected Vero cell monolayers, which were then washed, and bound antibody was detected using a commercial fluorescein isothiocyanate-conjugated antimouse antibody. The protocol was optimized using serially diluted known positive and negative sera, and is highly specific for antibody to orthopoxviruses. Cowpox virus is the only orthopoxvirus found naturally in Great Britain, but that the detected antibody was to cowpox virus was confirmed from 26 of these samples by polymerase chain reaction and nucleotide sequencing of the fusion protein gene, which enables differentiation of orthopoxvirus species (Chantrey et al. 1999). Only cowpoxvirus sequences were found (Blasdell 2006).

Vole density estimates for each primary session were calculated via the closed population model MTH in the program capture (Otis et al. 1978), assuming that no mortality or recruitment occurred between the first and final secondary trap sessions. Model MTH assumes heterogeneity in capture probabilities of individuals and differences in trapping probability between secondary sessions. The estimator of Chao & Lee (1991) was used to calculate vole abundance.

statistical analysis

Generalized linear models (GLMs) were run using S-plus Professional (Mathsoft Inc.). The fit of the models was assessed by Akaike's Information Criterion corrected for small samples (AICc) (Hurvich & Tsai 1989). This selects the most parsimonious model for the data by penalizing the better fit of more complex models according to the number of parameters included in the model. Models with a difference of AICc (ΔAICc) of less than 2 may be considered similar in their ability to account for the data (Sakamoto, Ishiguro & Kitagawa 1986). Generalized linear mixed models (GLMMs) were run using SAS (SAS/STAT 1992).

Calculating probabilities of cowpox infection

Raw data from the immunofluorescence assays characterized animals as being positive or negative for cowpox virus antibodies, or not having been caught in that primary session. Infection with cowpox virus results in long-term antibody production (Chantrey 1999) and therefore any negative results succeeding a positive result were considered to be the result of low titres and assumed to be positive (243 negative results succeeded positive ones from a total of 6061). From experimental studies, it is known that animals infected with cowpox virus develop an antibody response after approximately 2 weeks, and remain infected (and therefore infectious) for a period of 4 weeks from shortly after contracting the virus (Bennett et al. 1997; Chantrey 1999). Therefore, in a time series of antibody results, it could be assumed that an animal became infected, with uniform probability, between 14 days prior to its last negative result and 14 days prior to its first positive result. Time series of serological results were used to calculate probabilities that individual animals were susceptible to, infectious with, or recovered from cowpox virus infection for each trapping session: P(S), P(I) and P(R), respectively (see Telfer et al. 2002 for details). Captures in the first and last primary sessions were not included in the analysis, because they would have underestimated the prevalence of infectious individuals. Individual P(I), P(S) and P(R) values were summed for each trap session to estimate the total number of individuals infectious (It), susceptible (St) or recovered (Rt) at trap session tt.

For individuals seropositive at first capture, the estimated probability of infection was calculated according to their age. Juveniles (less than 6 weeks old) were assumed to have been seronegative 4 weeks previously (because they would hardly have been active outside the nest), and were therefore assigned a P(I) of 0·5 for that session (n = 515 of a total of 2706 individuals). Animals first caught positive that had already achieved adult weight (n = 884) were more problematic, as they may have been immigrants infected at an unknown time prior to dispersal on to the grid. Similarly, individuals seronegative at their last capture (n = 629) could have been infected prior to emigration from the grid or mortality, and could not confidently be assigned a probability of infection. For population-level analyses, it would be important to assign probabilities to these unknown captures (if ignored, the overall level of cowpox infection would be underestimated), but for the individual-level analyses of risk presented here, these individuals were simply not included in the analysis.

Investigating the risk of seroconverting

Factors influencing whether an individual animal seroconverted between trap sessions tt−1 and tt, were investigated using GLMMs with a logit link and binomial errors (as the risk of seroconversion was a binary measure). Individuals were classified as 1 or 0 at trap session tt, according to whether the individual seroconverted or remained antibody negative between trap sessions tt−1 and tt. Individuals were only included in the analysis if they were caught in both trap sessions concerned, and seropositive captures subsequent to the initial seropositive result were discarded. As explained above, adults caught first as antibody positive and recovered animals were not included in the analysis, while juveniles caught seropositive on first capture were classified as 1.

As is common with parasitological data, individual observations were nonindependent (individuals being sampled from the same site at the same time), and so site × trap-session was included as a random effect (Paterson & Lello 2003). The same individuals could also be sampled repeatedly over time, potentially leading to pseudo-replication. However, models were not run with individual identification number (id) as a random effect, as individuals appeared in the data only 1·38 times on average: fewer than the average number of primary captures due to seropositive captures after the first being discarded (range 1–7 and 73% seroconverted after only one trap session). Consequently, the variance component due to vole id was not estimable as a random effect in the GLMMs.

The analyses were conducted in three stages: first investigating whether month or season categories best accounted for seasonal patterns in risk, then investigating population-level covariates, and finally individual-level covariates. In the first stage, alternative models were not nested and stepwise selection procedures commonly utilized in GLMMs could not be used. Consequently, no random effects were included and seasonal GLMs were selected based on AICc. Four alternative season classifications were investigated: seas1 (months grouped in pairs), seas2 (months grouped in triplets), seas3 (months grouped into spring (March–April), summer (May–August), autumn (September–November) and winter (December–February) and seas4 (months grouped into two: May–October and November–April). Using the best seasonal model, the second and third stages included the site × session random effect and model selection followed a step-down procedure, eliminating two-way interactions first. Only variables significant at the 1% level were retained in the model, using SAS type III F-tests (SAS/STAT 1992) and denominator degrees of freedom calculated using Sattherwaite's formula (Littell et al. 1996).

Population-level covariates were those shared by voles caught in the same site on the same occasion, namely current or lagged vole density (1, 3 or 5 months), and interactions between season and these other variables. Starting with the best population-level model for seroconversion risk, individual-level covariates were then investigated: namely, sex, age [juveniles < 20 g, adults ≥ 20 g, and as a continuous variable (mass)], maturity and overall body-condition score. Two-way interactions between individual level covariates and between these and season and density were also included. It was particularly important to investigate whether any density dependence in the risk of seroconversion was caused by changes in the population structure (e.g. changes in the age (mass) or maturity of the population). This was assessed by including interactions between density and these individual covariates.

Juveniles seropositive on first capture could have been positive due to maternal antibodies rather than genuine infection with cowpox. A large number of such individuals could mask the true relationship between age or maturity and infection risk. Hence, the relationship between individual-level covariates and an individual's risk of becoming infected was first investigated using the full data set, including juveniles positive at first capture. Then, individual covariates were investigated using a reduced data set, excluding those juveniles most likely to have been seropositive due to maternal antibodies: those caught seropositive on first capture and seronegative subsequently (almost certainly maternal antibody), those caught positive on first capture with a mass of ≤ 14 g (more likely to be positive due to maternal antibody because of a lack of opportunity for infection), and those not caught again after the first positive capture (as their subsequent status could not be checked).

Finally, due to correlation between present and past density variables, the best model from stage 3 was used to investigate which density variable best fitted the data according to AIC model selection, with no random effects included. This was undertaken to confirm that the best density variable from the step-wise selection procedure was supported if a more rigorous model selection method was utilized.


In total, 2788 field voles were caught between May 2001 and July 2003 at the four trapping sites, with 5962 primary captures and 15 230 secondary captures of individual animals. The mean number of primary captures per individual was 2·14, and the maximum was 14.

field vole numbers

Population size was generally lower at KCS and PLJ than at BHP and ROB. Clear seasonal patterns in density were observed, with summer peaks and overwinter declines. Population size increased on all four grids throughout the 2·5-year study period (Fig. 1). The numbers of animals infected and susceptible also exhibited clear seasonal dynamics, increasing during the summer months with the influx of new juveniles and decreasing over winter (Fig. 2).

Figure 1.

Natural log density estimates per grid (0·3 ha) over the 2-year study period. Winter (October–March) is highlighted in grey. Population size estimates were calculated using the MTH model and Chao & Lee (1991) estimator using the program CAPTURE, except for January 2002 which was calculated using the minimum number alive method (Krebs 1966) due to the fact that this trap period only consisted of three secondary sessions. 95% confidence limits are plotted.

Figure 2.

Numbers of individual voles caught as infectious with, susceptible to or recovered from cowpox virus infection per trap session. Winter (October–March) is highlighted in grey. Trap sessions took place every 28 days except during the winter when there was a gap of 56 days between sessions. Numbers were calculated by summing the individual probabilities for each trapping occasion. These data are based on a reduced estimate of numbers because any last negative captures, or any first positive captures for adult animals are excluded.

cowpox prevalence and seroprevalence

Seroprevalence (the percentage of cowpox antibody positive samples per total samples per trap session) varied between 28% (KCS, July 2001, n = 31) and 100% (ROB, January 2002, n = 56 and March 2003, n = 65 and KCS, April 2003, n = 62). The proportion of animals infectious, susceptible and recovered varied seasonally, with highest infection levels occurring in summer and early autumn for all four sites (Fig. 3). The proportion of recovered animals remained high throughout the study, reaching levels of 0·98 (ROB, January 2002, n = 56).

Figure 3.

Proportions of animals caught per trap session that were infected, susceptible or recovered from cowpox virus infection for each trapping site over the 2-year period of the study. Winter (October–March) is highlighted in grey. Proportions were calculated by summing the respective probabilities and dividing by the total summed probabilities per trap session.

investigating the risk of seroconverting

In total, 1552 serological results from primary captures, representing 1133 individuals, were included in the analysis. Of these, 66·6% were recorded as seronegative for only one primary session before seroconverting.

In stage 1 of the analysis, season category seas2 provided the best fit to the data, but was no longer significant once density was included (F3,62·6 = 2·45, P = 0·071), and was eliminated once individual covariates had been fitted. In stage 2, seroprevalence risk was best modelled as increasing with density 3 months previously (F1,64·7 = 28·70, P < 0·001). Other density covariates were not statistically significant with this term included, but were significant if entered alone into the model, owing to obvious correlation between the different density terms. Density with a 3-month lag was confirmed as the best density variable using AIC in a GLM containing no random effects (ΔAIC = 13·5). GLMs were also run with more than one density term included, and a model with both current and lag-3 densities was found to be best (ΔAIC compared with lag-3 alone = 10·8), risk of seroconversion increasing with both. However, because only density with a 3-month lag had been retained in models including random effects, the effect of current density needs to be interpreted with caution.

In stage 3 (individual level covariates), body condition did not contribute to model fit, but males had a significantly higher risk of seroconverting than females (F1,1357= 25·86, P < 0·001), risk decreased linearly with mass on a logit scale (F1,1385 = 15·85, P < 0·001), and risk increased linearly on a logit scale with the density of voles 3 months previously (F1,64·7 = 14·54, P < 0·001) (Table 1). Mass as a continuous variable was a better predictor of risk than as a categorical age variable (juveniles < 20 g, adults ≥ 20 g) (F1,1393 = 10·45, P = 0·001), or sexual maturity (F1,1337 = 4·89, P = 0·027) alone, and neither age category nor maturity were significant if mass was included. There were no significant interactions between any of the model covariates. Predicted values of risk for males and females at a range of past densities and masses are shown in Fig. 4(a).

Table 1.  Parameter estimates (logit scale) for a model of the risk of an individual seroconverting between trap sessions, selected in a three-stage stepwise procedure starting with various season categories, then with a backward selection of population-level covariates, followed by a backward selection of individual level covariates (intercept represents adult males with a past density of 0)
EffectStandard estimateErrord.f.t-valuePr > |t|
intercept 0·5300·329106 1·610·111
dens3ago 0·0150·00465 3·810·000
Figure 4.

Predicted increase in risk of a male or female individual seroconverting in relation to (i) vole mass (g) with density 3 months ago held constant at 50 voles 0·3 ha−1, and (ii) vole density 3 months in the past for a vole of mass 25 g, for (a) model based on the full data set (including all juveniles), and (b) the reduced data set (not including juveniles caught first as seropositive and then subsequently seronegative, and any juvenile ≤ 14 g that was only captured first as seropositive). Dotted lines are 95% confidence limits on the fitted mean predictions.

The analysis of individual-level effects was repeated using the reduced data set, excluding juveniles caught seropositive on first capture and then seronegative subsequently (n = 55), or with a mass of ≤ 14 g (n = 75), or not caught again subsequently (n = 165). Seroconversion risk was again found to be higher for males (F1,1057= 22·46, P < 0·001), and to increase with the density of voles 3 months previously (F1,65·4 = 14·03, P < 0·001). However, the relationship between age (i.e. mass) and seroconversion risk was the reverse of that found for the full data set. Risk increased with mass (F1,1089 = 14·69, P < 0·001), and this was a better predictor of risk than age category (F1,1094 = 6·08, P = 0·014), or maturity (F1,1100= 5·68, P = 0·017) alone, and neither age nor maturity were significant in the model if mass was included. There were no significant interactions between any of the model covariates. Model estimates for the parameters (except mass) were very similar to estimates based on the full data set. Model estimates for males and females at a range of past densities and masses are shown in Fig. 4(b). In spite of this risk–mass relationship, the proportion of captures per mass category that were seroconversions (excluding those likely to be maternal antibody) was highest for the ‘young’ 10–19-g class and decreased with increasing mass (Fig. 5). This observed decline was significant: the frequency of seroconversion events in the different mass categories was significantly different from the frequency of capture events (χ2 = 176·24, d.f. = 4, P < 0·001).

Figure 5.

Proportion of total captures per mass category that were infected with cowpox virus (i.e. prevalence of infection as number of seroconversions per total number of capture events). Data did not include juvenile captures for which seropositive status is likely to have been due to maternal antibodies, and hence < 10 g has a value of 0 because all individuals in this age class were assumed to have been seropositive due to the presence of maternal antibodies.


Cowpox virus infection in field voles in Kielder exists at high prevalences, sweeping rapidly through the population with the influx of juvenile susceptibles in the breeding season. Seroprevalence levels (28–100%) were substantially higher than those in bank vole and wood mouse populations in Cheshire (0–72% and 0–26%, respectively) (Begon et al. 1999; Hazel et al. 2000). Estimates here, though, are similar to those from a cross-sectional study at Kielder that found a mean seroprevalence of c.75% (Cavanagh et al. 2004) but a range from 0 to 100% at different sites, possibly due to stochastic extinction of the infection at sites with low (post-crash) host densities not encountered in the present study.

For infection itself (not estimable in the study by Cavanagh et al. 2004), prevalence estimates here often exceeded 10%, whereas for bank voles and wood mice this level was only rarely reached in over 4 years (Hazel et al. 2000; Telfer et al. 2002). Furthermore, here a proportion of animals was infected at all times of the year, whereas for the bank voles and wood mice in Cheshire, prevalence during winter was very low and frequently zero (Begon et al. 1999; Hazel et al. 2000). Field voles may be more susceptible to cowpox infection than these other species, transmission may be more effective as a result of host behavioural differences or due to differences in virus survival outside the host, or infection may be longer lasting. Ongoing studies are aiming to distinguish between these possibilities.

This study was the first in which there was sufficient power to examine the relationship between individual level covariates and cowpox virus infection status in its rodent reservoir. As is often found, males were more likely to be infected than females. This may be due to behavioural differences between the sexes resulting from increased exposure of males to opportunities for transmission (for example, if juvenile males move to establish new territories), or to increased susceptibility to infection owing to testosterone levels suppressing immune function, as is widely observed in rodents generally (Alexander & Stimson 1988; Olsen & Kovacs 1996). The detailed basis of sex differences in this system, however, await further study.

The relationship between risk of infection and host age was complicated by the large number of juveniles that were seropositive at first capture. Laboratory observations indicate that maternal antibody to cowpox virus may be present in juvenile field voles of low mass (< 14 g) for 22–34 days after birth (n = 11), but that such individuals subsequently become seronegative (Blasdell 2006). Analysis of the full data set suggested that juveniles were more at risk of becoming infected than adults. However, once those individuals most likely to be carrying maternal antibody were removed from the analysis, risk increased significantly with age. The most likely explanation is that maternal antibodies were frequently detected, and that the genuine underlying pattern is an increased risk of cowpox virus infection with age. This occurred in spite of the fact that high prevalences of infection ensured that most individuals became infected when still relatively young. Juvenile/immature animals may have more restricted movements, or less frequent interactions with other individuals (copulation, aggressive encounters), that render them less at risk of infection. Mature animals also have higher levels of sex hormones, and higher levels of stress-related hormones associated with breeding and maintaining territories that may suppress immune function (McDonald & Taitt 1982).

This study demonstrates clearly that the risk of becoming infected with cowpox virus is positively correlated with host density 3 months previously. This corroborates the suggestion of a 3–6-month lag made by Cavanagh et al. (2004), inferred from snapshot seroprevalence data, and therefore potentially confounding the cumulative nature of seroprevalence and the numerical response of the pathogen. There was no significant interaction between maturity and density, nor between mass and density, and even when maturity or mass covariates were fitted, the additive effect of past density remained virtually the same and significant. Few studies have evaluated the role of parasites in host populations with cyclic dynamics, and most have found no clear evidence of a time delay (Hudson, Newborn & Dobson 1992; Moss et al. 1993). For example, helminth prevalence in bank voles exhibited direct, not delayed, dependence with host density (Haukisalmi, Henttonen & Tenora 1988; Haukisalmi & Henttonen 1990). However, Soveri et al.'s (2000) observation that a wide range of pathogens were prevalent in declining populations of rodents, and that of Descoteaux & Mihok (1986) of high seroprevalence for two viruses in declining rodent populations, are not inconsistent with a delayed response.

The delay between increases in host density and in the risk of infection is likely to be a property, in part at least, of the transmission process, which inevitably includes positive feedback. Initial increases in risk lead to increased numbers of infectious hosts, leading to further increases in risk, and so on, with the process being opposed and ultimately reversed by the diminished availability of susceptible hosts. The initial increase in risk with density may arise as a result of increased contact rates (and hence transmission) between infectious and susceptible hosts, even if the transmission is not ‘purely’ density-dependent (Begon et al. 1998, 2002). Alternatively, or additionally, the initial increase in risk may reflect an increased vulnerability to infection, possibly as a result of increased stress, aggressive encounters or reduced condition associated with high density.

For a parasite (or predator) to play a significant role in driving host density cycles, there must be a time delay between host density and enemy-induced changes in host population growth rate (May 1976). For this to occur, the parasite must satisfy three conditions. First, there must be a delay between host density and the risk of being infected, as demonstrated here clearly for the first time. Second, the pathogen must adversely affect the survival and/or reproduction of individual hosts. and third, this must translate into a significant effect at the population level (rather than being negligibly small or being counteracted by compensatory changes in other mortality factors). Studies of cowpox in bank vole and wood mouse populations indicate that cowpox infection substantially delays reproduction (and thus increases survival) during the summer, and causes reduced survival in winter (Telfer et al. 2002, 2005). The negative effect of cowpox virus infection on survival of field voles at Kielder appears to be considerably greater than this (Burthe 2005). There is therefore evidence for these populations that there is not only a delay in numerical response by a pathogen to host density, but that this pathogen also lowers host survival. Moreover, as most field voles at Kielder became infected when young at the high densities encountered in this study, the impact of infection on population growth rates could be especially significant. In voles, several studies have suggested that phase-related changes in juvenile survival rates and age at maturity may be important in driving population fluctuations (Gaines & Rose 1976; Getz et al. 1997; Oli & Dobson 1999, 2001; Getz, Simms & McGuire 2000; Ozgul, Getz & Oli 2004), including studies of field voles at Kielder (Ergon et al. 2001a,b). It can therefore be hypothesized that cowpox virus has delayed density-dependent effects on host dynamics. Furthermore, one key feature of microtine cycles is the low fitness of hosts for an extended period after host population density has crashed (Boonstra, Krebs & Stensteth 1998). This too is consistent with infection contributing to the population dynamics of the host, as infection rates will remain high for a period after a crash has occurred.

This study therefore emphasizes that the role of endemic pathogens, including cowpox virus, in affecting cyclic host dynamics needs to be explored more thoroughly. However, it is also important to recognize that infection risk fluctuated with overall host density 3 months previously, despite the fact that it is the density of susceptible hosts that drives infection dynamics, and that only susceptible hosts are vulnerable to infection and its possibly harmful effects. This could mean that the effect of cowpox on population growth rates is demographically negligible, due to the large numbers of recovered individuals in the population. Of course, this is no different from prey individuals maturing into an age class that is (relatively) invulnerable to predation. None the less, it is clear that the role of infection in affecting host dynamics cannot readily be deduced. Age/stage-based modelling of field vole populations at Kielder, parameterized using empirical data on infection prevalence and host demography, is therefore currently being undertaken to investigate this role. Only once this has been done will it be possible to move from a correlative study of past densities and present infection levels to a predictive study of present infection levels and future densities.


Work was funded by the Natural Environmental Research Council and licensed under home office project license PPL 40/1813. The Forestry Commission provided access to study sites. Gordon Brown, Jonathan Fairbairn, Matt Oliver, Laura Taylor, Gill Telford, David Tidhar, Rachel Yeates and many others provided valuable fieldwork assistance.