Temporal and between-site variation in helminth communities of bank voles (Myodes glareolus) from N.E. Poland. 2. The infracommunity level

SUMMARY The relative importance of temporal and spatial effects was assessed in helminth communities of bank voles (Myodes glareolus) in 3 woodland sites in N.E. Poland in the late summers of 1999 and 2002. Among common species the rank order of sites in relation to prevalence and abundance of infection was maintained between surveys. Site effects accounted for most of the deviance (in statistical models), and time was less important, so the exact location from which voles were sampled was of critical importance. The only exception was Syphacia petrusewiczi. In contrast, for derived measures such as species richness and diversity, most deviance was accounted for by host age, and the interaction between site and year was significant, implying that rank order of sites changed between years. Temporal effects on derived measures were generated primarily by a combination of relatively small changes in prevalence and abundance of the common, rather than the rare, species between the years of the study. In the medium-term, therefore, helminth communities of bank voles in N.E. Poland had a stable core, suggesting a substantial strong element of predictability.


I N T R O D U C T I O N
Helminth communities in hosts can be viewed at various levels of organization, ranging from the entire helminth fauna of a host taxon throughout its entire geographical range, through regional helminth faunas and component communities, to the infracommunities infecting individual hosts (Poulin, 1997 ;Behnke et al. 2008). Each level experiences different evolutionary and ecological pressures, but the degree to which these communities vary spatially and temporally has not been assessed adequately, even in such well-studied hosts as rodents. This may be surprising given the wealth of studies of rodent helminth communities, but few address this fundamental subject in sufficient detail (Kisielewska, 1970 a ;Haukisalmi et al. 1987 ;Montgomery, 1989, 1990 ;Tenora and Stanek, 1995). The issue of what a community is, and what structure, stability and predictability it might have, is crucial to establish before launching into complex theoretical treatments of communities.
In the preceding paper (Behnke et al. 2008) we found, counter to our expectations, that indices of regional and component community structure for helminths in the bank vole Myodes glareolus were remarkably dynamic temporally. This was surprising because we had expected greater stability at the higher levels of organization. Our analysis showed that the rank order of indices reflecting component community structure in 3 study sites in N.E. Poland was not maintained over a 3-year interval. We attributed this unexpected dynamism primarily to 3 helminths (Syphacia petrusewiczi, Mesocestoides lineatus and Cladotaenia globifera) each of which was rare, but which could achieve extremely high worm burdens in individual hosts, therefore contributing disproportionally to indices of parasite diversity.
We might expect more variation within relatively short-lived infracommunities, which change throughout the host's lifetime (Poulin, 1997). These are influenced by intrinsic factors such as host age because, generally, helminths are long-lived and the overall probability of the host being infected increases with age. Age effects have been quantified and replicated in many studies (Kisielewska, 1971 ; Montgomery and Montgomery, 1989 ;Behnke et al. 1999Behnke et al. , 2004Gregory et al. 1992 ;Bajer et al. 2005) and should now be included routinely when examining a sufficiently large sample of hosts spanning different age cohorts. Differences attributable to host sex are also documented, but are less consistent and not always replicable for specific host-parasite combinations (Kisielewska, 1970 b ;Poulin, 1996 ;Behnke et al. 2001 ;Ferrari et al. 2004 ;Eira et al. 2006).
Extrinsic factors such as season, year of study, site, environment etc. have also been assessed to lesser or greater extents (Lewis, 1968 ;Langley and Fairley, 1982 ;Haukisalmi et al. 1988 ; Montgomery and Montgomery, 1989 ;Abu-Madi et al. 2000). However, seldom has the contribution of any specific quantified factor been evaluated in a manner that then allows its importance to be clearly established relative to other factors. Which factors are relatively the most important, and which the most predictable, in terms of their influence on infracommunity structures ? Do summary measures of infracommunity structure show stability with time in different host populations and, if so, are the differences maintained between surveys ? How do common versus rare species contribute to infracommunity structure ? Our earlier work showed some temporal stability in helminth infracommunities in bank voles from one of our study sites (Urwitałt ; see Bajer et al. 2005) leading us to predict that, with age and sex effects taken into consideration, differences between sites would be a major factor in explaining variation in our measures of infracommunity structure, whereas time would be relatively less important. In this paper we test this idea.

Collection of voles
In this paper we refer to bank voles as Myodes glareolus following the recent recommendations that Myodes rather than Clethrionomys is the valid name for this genus of rodents (Carleton et al. 2003 ;Wilson and Reeder, 2005). Methods used for trapping rodents, and for sampling and processing trapped animals have all been fully described by Behnke et al. (2001). Age categories were established as described by Behnke et al. (2001). The breakdown of voles sampled by year, site and age is given in the accompanying paper (Behnke et al. 2008).

Measures of community structure
Infracommunity structure was assessed following Kennedy and Hartvigsen (2000) by mean number of helminth species per vole, maximum number of helminths and species density distribution across the sample, mean number of helminth individuals per vole, mean Brillouin's Index of Diversity (ID) per vole (infected and uninfected), maximum Brillouin's ID, mean abundance and prevalence of individual species. Brillouin's index of diversity is considered to be the most appropriate for fully censussed communities such as those of the alimentary tract (Kennedy et al. 1986 ; but see also Washington, 1984). In some analyses, we also excluded the rare species whose overall prevalence was less than 10 %, in order to specifically evaluate their relative contribution to infracommunity structure.

Statistical analysis
The statistical approach adopted has been documented comprehensively in our earlier publications (Behnke et al. 2001 ;Bajer et al. 2005). We used general linear models (GLM) with Poisson errors (for species richness) or normal errors (for Brillouin's index of diversity) implemented in R version 2.2.1 (R Core Development Team). Full factorial models that converged satisfactorily were simplified using the STEP procedure and tested for significance using deletion of terms beginning with the highest order interaction by comparing models with or without that interaction. Changes in deviance were interpreted with Chi 2 (x 2 ) (Poisson errors), or F (Gaussian errors) tests respectively. Minimum sufficient models were then fitted (all significant interactions and main effects+any main effects that featured in interactions) and the process was repeated to obtain values for changes in deviance, test statistics and probabilities. The percentage of deviance accounted for by each significant main effect or interaction was calculated as recommended by Xu (2003) using the equation : percentage deviance explained (by main effect of factor or interactions between factors) where V o is the deviance of the null model with no factors or interactions fitted, V 1 is the deviance of the minimum sufficient model with factor or interaction under evaluation fitted and V 2 is the deviance of the minimum sufficient model without the factor or interaction under evaluation.
Species density distributions, representing the frequency distributions of infracommunity species richness, were tested for goodness of fit to different distributions as explained by Behnke et al. (2001).
All distributions were tested for goodness of fit by x 2 .
Prevalence data (percentage of animals infected) are shown with 95 % confidence limits, calculated as described by Rohlf and Sokal (1995) employing bespoke software and analysed as described previously (Behnke et al. 2001(Behnke et al. , 2008 employing maximum likelihood on log linear analysis of contingency tables using the software package SPSS (Version 12.0.1.).
Summary figures for parasite abundance are expressed as arithmetic means¡S.E.M. These means reflect the abundance of infection as defined by Margolis et al. (1982) and Bush et al. (1997) and include all subjects within the specified group, infected and not infected, for which relevant data were available.
The degree of aggregation in the data was calculated by the index of discrepancy (D) as described by Poulin (1993) and the index of dispersion (I, variance to mean ratio). Frequency distributions of individual taxa were also tested for goodness of fit to negative binomial, positive binomial and Poisson models by x 2 as described by Elliott (1977) and the negative binomial exponent k is given as appropriate.
Abundance data were analysed by fitting GLMs with negative binomial errors, using the glm.nb routine in the MASS library of R (Venables and Ripley, 1997) and, if they converged, simplified as described above and the likelihood ratio statistic (LR) is given. When models in R with negative binomial errors did not converge, we first explored simpler models. In the case of H. mixtum we examined a model with just main effects, and since host sex was not significant we removed this and proceeded further by fitting a full factorial model with all three remaining factors. For analysis of A. tetraptera, models with sex and age would not converge, so the preliminary analysis was conducted fitting a model with just site and year plus their interaction. In the case of S. petrusewiczi, no models with interactions would converge. In both cases, A. tetraptera and S. petrusewiczi, we used a Box-Cox transformation to normalize the data and fitted models with Gaussian error structures. For all models with negative binomial errors we give the LR statistic and for those with Gaussian errors F is provided.
When the assumptions of parametric tests were not satisfied, the non-parametric Mann-Whitney U-test was used for 2 group comparisons, the 1-way Kruskal-Wallis ANOVA when more than 2 groups were involved.
We used canonical discriminant function analysis (CDF) in SPSS as an additional approach to evaluating the relative importance of the influence of site and year on parasite burdens. Quantitative parasite data for each of the 13 species of helminths were first standardized (by log e (x+1) transformation of individual worm burdens for each species, then subtraction of mean log e value for each species and division by the standard deviation) before analysis.

Mean species richness
The overall mean number of species harboured per host (all voles combined) was 1 . 48¡0 . 058 (variance to mean ratio=0 . 80). There was no significant effect of year (Table 1 ; Minimum sufficient ANOVA with Poisson errors, main effect of year, x 1 2 =0 . 06, P=NS), the values being very close in both years (1 . 44 in 1999 and 1 . 50 in 2002). However, there was a significant difference between sites and this varied between the two years (Table 1, 2-way interaction, year * site, x 2 2 =16 . 6, P=0 . 0002 and main effect of site, x 2 2 =9 . 4, P=0 . 009). Mean species richness increased at Urwitałt by 25 . 8 % and was the highest of the three sites in 2002. It similarly increased at Tałty by 25 . 0 %, but fell at Pilchy by 32 . 7 %, making this site the poorest in terms of mean species richness in 2002. Thus, the interaction between year and site explained more of the deviance than either factor alone ( Table 2).
In contrast to the role played by extrinsic factors, here the intrinsic factor host age was considerably more important in explaining deviance (Table 2), mean species richness increasing with age ( Fig. 1A ; main effect of age, x 2 2 =43 . 1, P<0 . 0001). Moreover, there was an indication that the rate of increase with age varied significantly between years (borderline significance for 2-way interaction year * age x 2 2 =5 . 7, P=0 . 058), being relatively slower and spanning a lower range in 1999 compared with 2002 ( Fig. 1A).
When all the rare species (those with an overall prevalence of less than 10 %, including S. petrusewiczi, T. muris, A. annulosa and all cestodes except C. henttoneni), were excluded from these analyses, the values changed only marginally. The largest change to mean species richness was at Urwitałt where exclusion of the rare species resulted in a fall of 18 . 4 % in 2002 (Table 1). The largest change between years was at Tałty, now a 46 . 6 % increase (rather than 24 . 6 % increase when all species were included). Overall mean species richness fell by 13 . 5 % to 1 . 28.

Infracommunity diversity
The maximum number of helminth species per vole ranged from 2 in 1999 in Tałty to 5 in Urwitałt in 2002 (Table 1), and the exclusion of rare species made little difference to these values.
The mean number of helminths (all species combined) harboured per vole underwent a major change, falling overall from a mean of 75 . 7 to just 13 . 6 in 2002 (Table 1 and 82 % reduction ; minimum sufficient model with negative binomial errors, main effect of year, LR 1 =67 . 9, P<0 . 001). This year effect accounted for more of the deviance than any of the other terms in the model (Table 2) and was particularly evident in Tałty (Table 5, 91 . 9 % reduction) and Urwitałt (81 . 9 %), but less so in Pilchy (34 . 1 %). However, there was also a significant 2-way interaction with site (site * year, LR 2 =18 . 2, P=0 . 0001) and the main effect of site was significant (LR 2 =7 . 3, P=0 . 025), but both accounted for less explained deviance than the year effect ( Table 2).
The analysis also revealed a highly significant main effect of age (LR 1 =35 . 8, P<0 . 001). This was evident in both years ( Fig. 1B) but the differences between age classes were greater in 2002 compared with 1999, although not significantly so (no significant 2-way interaction between year and age). The model also had a significant 2-way interaction between site and age (LR 4 =28 . 5, P<0 . 0001) but no clear pattern emerged when this was scrutinized carefully (not illustrated).
When the rare species were removed from the analysis, the values for the mean number of helminths harboured dropped enormously at both Urwitałt and Tałty but less so at Pilchy, and this was mainly accounted for by the removal from the dataset of S. petrusewiczi. Without the rare species, some of which can generate huge numbers of individuals when present, the values were more stable across sites and between years. In both years the mean number of helminths was highest at Pilchy but because of a reduction at Tałty between 1999 and 2002, Tałty fell from intermediate position to bottom between the surveys. In contrast note the stable values at Urwitałt ( Table 1).
Analysis of Brillouin's index of diversity (ID) in R (with Gaussian errors) yielded a model indicating that age explained the greatest proportion of deviance ( * For these analyses we excluded all species for which prevalence was less than 10 %. Therefore the analysis was based on 5 species which were H. mixtum, H. glareoli, A. tetraptera, M. muris and C. henttoneni. F 2,352 =5 . 5, P=0 . 004) and these changed between the years of the study (2-way interaction site * year, F 2,350 =13 . 7, P<0 . 0001). In 1999 the lowest index was at Tałty and the highest at Pilchy, but by 2002 the index had increased at both Urwitałt and Tałty and dropped in Pilchy. There was no overall difference in Brillouin's ID between the sexes (males=0 . 19¡0 . 019, females= 0 . 20¡0 . 021).
Exclusion of the rare species (Table 1) made very little difference to the mean values of Brillouin's ID recorded for the three sites and in both years. Overall (sites and years combined) it fell by 15 % from 0 . 2 to 0 . 17, and the range of change was from 5 . 3% to 23 . 5 %. As with mean species richness the biggest change was recorded at Urwitałt.

Species density distributions
In the combined data-set and in each year, with sites combined (Fig. 2B), the modal class was just 1 species of helminth (35 . 2-40 . 3 % of voles). This was also the case for 4 of the 6 subsets of data ( Fig. 2A), but in 1999 at Pilchy the modal class was 2 species of helminths (39 . 7 %) and in 2002 at Urwitałt it was 3 species (26 . 4 %). Four-species infections were rare in 1999 (only 1 vole from Pilchy), but were present in all three sites in 2002. The difference between the distributions in 1999 and 2002 had borderline significance (with sites combined ; x 5 2 =10 . 4, P=0 . 06). In 2002 there were relatively more (in percentage terms) uninfected voles, fewer with 1 and 2 species of helminths but more with 3, 4 and 5 helminth species. The combined data-set ( Fig. 2C) did not differ significantly from the null model for interactions of parasite species in an assemblage (Janovy et al. 1995 ; x 4 2 =2 . 11, P=0 . 83). Table 3 shows that by 2002, overall prevalence of helminths had dropped by 7 . 1 % (from 85 . 6 to 78 . 5 %), but the extent of the change between the two surveys varied between sites (year * site * presence/ absence of helminths, x 2 2 =11 . 1, P=0 . 004). Thus, whereas at Urwitałt and Pilchy prevalence of helminths dropped between 1999 and 2002 (9 . 4 and 20 . 1 %, respectively), at Tałty it actually increased by 12 . 3 %. The prevalence of helminths also increased significantly with increasing host age ( Fig. 3A ; Table 2. Percentage of variation in data (deviance) explained by extrinsic and intrinsic factors affecting the measures of infracommunity structure and diversity, and the abundance of helminths (In each case the output from the most parsimonious and appropriate minimum sufficient model is given. Thus, only the significant main effects and interactions, and non-significant main effects if a component of one of the interactions, have been included. Models for total helminth burden and individual species are models with negative binomial error structures unless stated otherwise below. For further details of the statistical models, see the text. Note that some 2-way and 3-way interactions and the 4-way interaction are not given because these were not significant.) age * presence/absence of helminths, x 2 2 =29 . 4, P<0 . 0001).

Source
The overall reduction in prevalence of total helminths between the two surveys was mainly accounted for by nematodes (an overall reduction of 12 . 5 %), falling at both Urwitałt and Pilchy but staying virtually identical at Tałty (Table 3 ; year * site * presence/absence of nematodes, x 2 2 =8 . 9, P= 0 . 012). For nematodes there was also a highly significant increase in prevalence with host age ( Fig. 3B ; age * presence/absence of nematodes, The greatest overall percentage change between the years was observed with H. glareoli, the second most prevalent species in 1999, with prevalence falling overall from 36 . 0 % in 1999 to 10 . 5 % in 2002. This drop in prevalence is apparent in Fig. 3C, but as can be seen it was confounded by variation between the sexes and age classes (year * sex * age * presence/ absence of H. glareoli, x 2 2 =7 . 9, P=0 . 020). There was also an independent marked difference between sites (Table 3, site * presence/absence of H. glareoli, x 2 2 =123 . 7, P<0 . 0001), with prevalence being relatively high at Pilchy in both years and comparatively low at the other two sites. So at this level, the site effect was maintained between the two years of the study. However, because essentially this species was largely encountered at just the 1 site, in a second step the analysis was confined to voles from Pilchy, where overall prevalence was 50 . 8 %. The difference between years was now highly significant (year * presence/absence of H. glareoli, x 1 2 =35 . 5, P< 0 . 0001) and no other factors came into the equation. Prevalence at Pilchy was marginally higher in females ( H. mixtum was the most prevalent parasite in 1999, but not in 2002, when A. tetraptera was more common (Table 3). The strongest effect on the prevalence of H. mixtum was host age, prevalence increasing significantly with increasing age ( Fig. 3D ; age * presence/absence of H. mixtum, x 2 2 =29 . 1, P< 0 . 0001). Overall the prevalence of H. mixtum fell by 7 . 9 % between the two surveys. The reduction was most marked at Urwitałt (31 . 1 % reduction), but was also evident at Tałty (9 . 7 %), and this difference between sites and the degree to which prevalence changed at each was significant (Table 3 ; year * site * presence/absence of H. mixtum, x 2 2 =6 . 6, P=0 . 0369). H. mixtum was not encountered in either year at Pilchy. There was no significant difference in prevalence between sexes (males=36 .  (Table 3). At Pilchy, which had shown the highest prevalence in 1999, there was little change and it was still the site with the highest prevalence of this species. However, as can be seen from Table 3, at both of the other sites prevalence increased quite markedly (year * site * presence/ absence of A. tetraptera, x 2 2 =20 . 1, P<0 . 0001) but overall the relative ranking of sites was maintained across the two years. There was also a significant increase in prevalence with host age (Fig. 3E Mastophorus muris was another species that appeared to increase in prevalence, this time at all three sites, although not to the same degree (Table 3). When other factors were taken into account the year effect was not significant. However, there was a highly significant effect of site (Table 3 ; site * presence/absence of M. muris, x 2 2 =38 . 1, P<0 . 0001). The highest prevalence in 1999 was at Pilchy but in 2002 prevalence was higher at Urwitałt, where voles showed a 23 . 3 % increase between the two surveys. At Tałty this species continued to be relatively rare, so between the two surveys there was a reversal of ranking among the two sites where the parasite was common. Prevalence increased significantly with host age, with a particularly steep rise between age classes 2 and 3 ( Fig. 3F (Table 3), but the change was the greatest at Tałty (30 . 3 %), where in 1999 prevalence of this species had been the highest. There was no significant difference in prevalence between the sexes (males=6 . 5% [2 . 9-13 . 0] and females=7 . 5% [3 . 6-14 . 2]).
The remaining nematodes were rare. A. annulosa was only detected in 2002 at Urwitałt and Tałty but not at Pilchy and, interestingly was recovered only from 6 female voles of age class 3. Trichuris sp. was only recovered from 1 vole in 1999 from Pilchy.
In contrast to nematodes, the overall prevalence of cestodes increased by 17 . 7 % (larval cestodes 8 . 5% and adult cestodes 17 . 9 %) between 1999 and 2002, the increase occurring mainly in voles from Urwitałt and Tałty (Table 3). This difference between years and sites was significant but was also confounded by host sex (year * site * sex * presence/absence of all cestodes, x 2 2 =7 . 9, P=0 . 019), as illustrated in Fig. 4A. The site and sex effect was also complicated by host age (Fig. 4B ; site * sex * age * presence/absence of all cestodes, x 4 2 =16 . 5, P=0 . 0024), an age-dependent increase in cestode infections being clearly apparent among male voles at Urwitałt and Pilchy but not at Tałty. Despite the variation between age classes, there was no clear pattern among female voles.     Note 1 vole in 2002 had nematode larvae in its liver that could not be identified, but were considered to be a different species to those listed above.
At Urwitałt the prevalence of both adult and larval cestodes increased between the two surveys, whereas at Tałty the increased prevalence in 2002 was mainly accounted for by adults (Table 3). At Pilchy, the prevalence of both larvae and adults was much the same in both years.
The remaining cestodes were not analysed further because of their low prevalence, but a few subjective observations are justified. P. omphalodes was present at Urwitałt in both surveys with a similar prevalence and absent on both occasions at Pilchy. It was found in a single vole at Tałty in 2002 but not in 1999. M. lineatus was present in all three sites and showed a very similar low prevalence in both years. T. martis was absent from Tałty, showed a similar prevalence at Pilchy in both surveys, and appeared at Urwitałt in 2002, where it had not been detected in 1999. In contrast, T. mustelae showed similar prevalence at both Urwitałt and Pilchy in both surveys, and appeared at Tałty in 2002. C. globifera was not detected at all in 1999, but was present in 1 vole each at Urwitałt and Pilchy in 2002.

Frequency distributions and measures of aggregation
Of the 14 species of helminths recovered, only 5 species (4 nematodes and 1 cestode) had an overall prevalence greater than 10 % (Table 3) and quantitative analyses were confined to these species, with S. petrusewiczi. The latter was included because in some subsets of data it showed high prevalence, even though its overall prevalence was only 7 %, and because the Berger-Parker dominance index identified this species as dominant more often than others (accompanying paper, Behnke et al. 2008). Frequency distributions were fitted to all species for which quantitative data were available, by site, by year and in relevant combinations, and were tested for goodness of fit to the positive and negative binomial and Poisson distributions. As can be seen from Table 4, the parameters calculated mostly indicated good approximations to the negative binomial. In most cases D was well above 0 . 6. The exceptions were H. mixtum in 1999 at Urwitałt, the combined data-set for this site (but in the latter case a test for goodness of fit to the negative binomial failed to reject this distribution) and H. glareoli in 1999 at Pilchy. With the single exception of H. glareoli at Urwitałt in 1999 and in the combined data-set for that site, the index of dispersion was greater than or equal to 1, and mostly considerably higher, ranging to 3945 for S. petrusewiczi in 1999 at Urwitałt.

Abundance of infection
Analysis of the abundance of H. mixtum was conducted on a model excluding sex, because when we included sex the models would not converge. A model including all the main effects without interactions, indicated that sex was not significant (model with negative binomial errors, main effect of sex, LR 1 =0 . 04, P=0 . 85). Table 3 shows that H. mixtum was totally absent from Pilchy in both years and Table 5 shows that abundance was higher at Urwitałt compared to Tałty in both years (main effect of site, LR 2 =187 . 7, P<0 . 0001). Abundance fell significantly between 1999 and 2002 (main effect of year, LR 1 =21 . 6, P<0 . 0001), but proportionally the reduction was greater at Urwitałt, where infections were heavier in 1999 (Table 5 ; 2-way interaction site * year, LR 2 = 10 . 7, P=0 . 005).
There was a significant main effect of age (LR 2 =19 . 8, P<0 . 0001), abundance of H. mixtum increasing with age (mean worm burdens in age class 1=0 . 9¡0 . 34, age class 2=2 . 2¡0 . 28 and age class 3=2 . 9¡0 . 38), but this was confounded by a significant 2-way interaction with year of study (LR 2 =17 . 3, P=0 . 0002). Since no voles at Pilchy carried this species, the interaction between year and age is illustrated only for voles from the other two sites (Fig. 5A). This shows that abundance was consistently higher in 1999 and that in relative terms, despite lower overall abundance in 2002, the rate of  increase from age class 1 to 3 was greater than in 1999.
Statistical analysis of the abundance of H. glareoli yielded a simple model with just 2 highly significant terms. There was a significant difference between sites (Table 5, LR 2 =151 . 2, P<0 . 0001) with this species being predominantly encountered at Pilchy, and only occasionally occurring in voles from the other two sites. This was the case in both years, although in 2002, overall abundance fell quite markedly (main effect of year, LR 1 =29 . 5, P<0 . 0001).
M. muris was virtually absent from Tałty in both years, only 1 worm being detected in 2002, so the site effect was highly significant ( Table 5 ; main effect of site, LR 2 =35 . 2, P<0 . 0001). Abundance was similar at Urwitałt and Pilchy (Table 5), and whilst in both it increased from 1999 to 2002, when other factors were taken into consideration this between-year change was just outside our cut-off for significance (main effect of year, LR 1 =3 . 2, P=0 . 073). To some extent it was confounded by a significant 2-way interaction (year * site, LR 2 =9 . 5, P=0 . 008), arising mainly because at Tałty the increase was from zero in 1999 to just 0 . 01 in 2002, whilst at Urwitałt and Pilchy abundance increased more markedly (Table 5, 9 . 3and 3 . 5-fold increases, respectively).
There was a highly significant main effect of age (Fig. 5B, LR 2 =21 . 5, P<0 . 0001), abundance increasing with age in both years although the relative increase was more prominent in 2002 (but the 2-way interaction year * age was just outside significance LR 2 =5 . 71, P=0 . 058). There was also a significant 2-way interaction site * age (not illustrated, LR 4 =10 . 3, P=0 . 035). In this species abundance is usually higher in females (main effect of sex, LR 1 =9 . 85, P=0 . 0017), and this was the case in both years (Fig. 5D in 1999 and 2002 mean worm burdens were 3 . 6 and 4 . 9 times higher, respectively in females compared with males). Fig. 5C, based only on data from Urwitałt and Pilchy, where the parasite was most prevalent, shows that the abundance of M. muris was already marginally higher in female age class 2 voles, but was highest in the oldest animals. There were no higher order interactions in this case.
For S. petrusewiczi, neither full factorial models with negative binomial error structures nor simpler models with only main effects converged satisfactorily in R. Therefore, we used a Box-Cox transformation of the data and fitted a model with a normal error structure (Box-Cox lambda¡CL= x4 . 22450¡x4 . 20088 & x4 . 24813 ;Likelihood= 581 . 434). Abundance varied significantly between sites and as Table 5 shows S. petrusewiczi was rare in both years at Pilchy but more abundant in both at Urwitałt and Tałty (main effect of site, F 2,353 =7 . 19,  (Poulin, 1993). 4. Not possible to calculate parameters (too few infected animals). * Not significantly different from the negative binomial. " Significantly different from the negative binomial. Note that where no symbol is shown, it was not possible to test for significance because of too few degrees of freedom, or too few voles were infected. P<0 . 001). However, abundance dropped markedly between 1999 and 2002 in all sites (main effect of year, F 1,353 =19 . 0, P<0 . 001) and the relative ranking of Urwitałt and Tałty was reversed (2-way interaction between site * year, F 2,351 =11 . 4, P<0 . 001). Abundance did not differ significantly between age classes nor between sexes. The main effects of year and site were also confirmed by non-parametric tests (Mann-Whitney U test on year, z=x3 . 9, P<0 . 001 and Kruskal Wallis test on site, x 2 2 =11 . 3, P=0 . 004) but no age nor sex effects were evident.
Attempts to fit full factorial models with negative binomial errors to the abundance of A. tetraptera (dependent variable) did not converge in R. More restricted models with age, or sex and interactions with age and/or sex, all failed to converge. Therefore we fitted a model with just site and year as main effects and their interaction. As Table 5 shows, in both years abundance of A. tetraptera was highest at Pilchy and lowest at Urwitałt (model with negative binomial errors, main effect of site LR 2 =14 . 3, P=0 . 0008). There was also a reduction in abundance between 1999 and 2002 in two sites but a relatively small increase in abundance at Urwitałt (2-way interaction site * year, LR 2 =31 . 4, P<0 . 0001).
Further support for this interpretation came from a 2-way non-parametric analysis of variance, which also supported the strong site effect but also gave a weaker 2-way interaction and a borderline year effect (main effect of site, H 2 =38 . 1, P<0 . 0001 ; main effect of year H 1 =4 . 12, P=0 . 04 ; site * year, H 2 = 10 . 2, P=0 . 006).
To eliminate the possibility of confounding effects from sex and age, we also fitted a model in R using Box-Cox transformed data, and with a normal error structure (Box-Cox lambda¡CI=x1 . 01858¡ x1 . 01052 & x1 . 02665 ; Likelihood=x128 . 248). This confirmed the effects of site on A. tetraptera (F 2,352 =36 . 5, P<0 . 0001) and the interaction between site and year (F 2,352 =8 . 12, P<0 . 0001) but indicated that the main effect of year was also significant (F 1,352 =9 . 3, P=0 . 0024). Furthermore, the model based on Box-Cox transformed data indicated that abundance varied significantly with age (main effect of age F 2,352 =7 . 13, P<0 . 001) but no clear consistent pattern was evident (Fig. 6). The mean worm burden was highest in age class 1 in 3 data sets (Tałty in 1999, Pilchy in 1999and 2002, in age class 2 in 1 data set (Tałty 2002) and in age class 3 in 1 data set (Urwitałt 2002). The significant main effect of age was primarily attributable to the abundance of A. tetraptera being more than twice as high in age class 1 voles (8 . 5¡3 . 74) compared with older animals (age class 2=3 . 0¡0 . 57 ; age class 3=3 . 6¡1 . 08). The strongest effect on C. henttoneni was site (model with negative binomial errors, main effect of site LR 2 =47 . 8, P<0 . 0001), and as can be seen from Table 5 abundance was highest at Urwitałt and lowest at Pilchy ; this was consistent in both years. There was a relatively smaller but significant effect of year (main effect of year LR 1 stat=21 . 4, P<0 . 0001). Abundance increased in all sites in 2002, most prominently at Urwitałt and least at Pilchy, although proportionally relative to 1999 the increase was greatest at Tałty (x8.2-fold), compared with Urwitałt (x5.25-fold) and Pilchy (x3.5-fold).
None of the remaining parasites were present in sufficient animals to merit analysis. Table 2 summarizes for each statistical model the percentage of variation (deviance in R) explained by each of the significant terms. It is clear from these data that the major source of variation, for the 5 species that were examined quantitatively, arose from the site effect. In 4 of these 5 species there was also a significant time effect but this accounted for a considerably smaller percentage of explained deviance. The exception was S. petrusewiczi where the year effect was much stronger, as was the interaction between site and year, with a resultant change in the relative ranking of sites between the two surveys. Intrinsic factors generally played a lesser role in explaining variation in parasite burdens.

Canonical discriminant function analysis
Canonical discriminant function (CDF) analysis revealed 5 axes that cumulatively accounted for all of the discrimination. Axis 1 (Eigenvalue=1 . 94) accounted for 66 . 6 % and Axis 2 (Eigenvalue=0 . 524) accounted for a further 18 % of the discrimination, and therefore we did not examine Axes 3-5. Axis 1 (Table 6 and Fig. 7) contrasts H. glareoli and A. tetraptera with H. mixtum and C. henttoneni, and can be interpreted as reflecting differences between sites. Both H. glareoli and A. tetraptera were most abundant at Pilchy, and least at Urwitałt. Both H. mixtum and C. henttoneni were most abundant at Urwitałt and least at Pilchy. Axis 2 contrasts C. henttoneni, M. muris and T. martis against H. glareoli and H. mixtum. As can be seen from Table 3   H. glareoli declined. Therefore, this axis can be regarded as a reflection of the major changes with time, i.e. between the surveys. It is clear from the distribution of symbols and the centroids for each of the 6 subsets of data (3 sites and 2 years) that, to a large extent, the site effect was maintained between 1999 and 2002 (compare sequence of centroids labelled 1, 2 and 3, with the sequence 4, 5 and 6). In terms of the year effect, the centroids of all three sites have moved downwards (more negative) on axis 2, with the greatest change at Urwitałt (centroids 1 and 4) and the least at Tałty (centroids 2 and 5). Overall therefore the relative order of the site effect (from left to right in Fig. 7) in both years was the same, although the difference between sites was smaller in 2002 (a slight contraction towards the centre) than in 1999.

D I S C U S S I O N
The key objective of this work was to test for relative constancy of the infracommunity through time, in the face of spatial differences among sites : this is exactly what we found. In marked contrast to measures of regional helminth fauna and component community structure which were fluid, changing from year to year and altering in rank order between sites (see accompanying paper, Behnke et al. 2008), infracommunity structure was clearly dominated by the site effect when examined at a variety of levels (e.g. prevalence, aggregation, abundance etc.) for individual parasite species that were common and which could be treated quantitatively. In general, relevant measures showed the same rank order in 1999 and 2002 across sites. When all the abundance data were combined, standardized and analysed by Canonical Discriminant Function Analysis, year and site effects were clearly reflected in the two major axes that collectively explained 84 % of the discrimination. Site was relatively more important in explaining 66 % of total discrimination and remained relatively stable with time, with year of study explaining only 18 %. This finding is consistent with earlier studies, which have generally shown that the greatest source of variation in quantitative data (abundance) fitted to statistical models derives from differences among sites Montgomery, 1989, 1990 ;Eira et al. 2006) rather than to changes in time (Kisielewska, 1970 a ;Keymer and Dobson, 1987 ;Bajer et al. 2005 ;Bugmyrin et al. 2005). Therefore, at the level of infracommunity structure, there seems to be considerable medium-term stability in these communities and, as in other helminth and protozoan infections, the site from which animals are taken is critically important (Mollhagan, 1978 ;Thul et al. 1985 ;Calvete et al. 2004 ;Booth, 2006). The haplodiploid oxyurid nematodes can be singled out as the most problematic in studies of helminth communities and this is really because the tools we have to analyse communities cannot deal with their highly aggregated distributions. Helminth parasites are known to show aggregated distributions among hosts (Anderson and May, 1978 ;Haukisalmi, 1986 ;Shaw and Dobson, 1995). However, in the Oxyuridae in general, and in Syphacia spp. in particular, aggregation can be dramatic. In the case of S. petrusewiczi in this study, the overdispersion was extreme (see also Kisielewska, 1970 b): so extreme that models based on negative binomial error structures would not converge satisfactorily. Not only were infections aggregated in relatively few voles, but the abundance and prevalence of S. petrusewiczi varied more in respect of year than of site, distinguishing this species from all the others for which it was possible to analyse quantitative data. Because worm burdens could be so high, S. petrusewiczi had a disproportional influence on the derived statistics to which it contributed, particularly those concerned with regional helminth fauna and component communities (see accompanying paper, Behnke et al. 2008). The influence of this species on infracommunity statistics was far less substantial and essentially confined to analysis of total helminth The correlation between each species and functions 1 and 2 is given in burdens. Two other relatively uncommon, highly aggregated species were the larval stages of the cestodes M. lineatus, and C. globifera (see Behnke et al. 2008). Neither contributed significantly to infracommunity structure because their prevalence was so low. A key finding at infracommunity level that remained stable between the two surveys was the total absence of H. mixtum from Pilchy and the presence instead of more intense infections with H. glareoli, relative to the other two sites. A third survey (2006) has confirmed this finding (Behnke et al. unpublished data). Although the prevalence and abundance of H. mixtum were lower in 2002 in both Urwitałt and Tałty, the relative ranking remained the same as in 1999, with infections being more abundant and prevalent at Urwitałt in both years. Similarly, although the prevalence of H. glareoli fell between 1999 and 2002, it remained relatively high at Pilchy in both years. Here also, quantitative analysis revealed that the site effect explained the greater proportion of deviance in the data, and although significant, the year effect was relatively minor in comparison.
A similar pattern was shown by M. muris. This species was absent from a different site, Tałty in 1999, and only one worm was collected here in 2002. It increased marginally at Pilchy but substantially, over 3-fold, at Urwitałt, and the figures for 2002 also contrast strongly with data collected in 1998-2000 that reported a very low prevalence at Urwitałt for this species (Bajer et al. 2005). Not surprisingly, the site effect was most important and the betweenyear effect was considerably weaker. Interestingly, M. muris was a species that was clearly subject to some degree of regulation by host sex, and unusually the bias was in the direction of female hosts. The related spirurid nematode Protospirura muricola, which also lives in the stomach of its host Acomys dimidiatus, showed a similar, but non-significant bias in favour of female mice (Behnke et al. 2004). Our finding of a female host bias for M. muris are supported by the data of Haukisalmi and Henttonen (2000) who also found the prevalence of M. muris higher in female bank voles in August and September (the same months as our surveys) in Finland. The basis of the dichotomy between the sexes in our study is not known, but may be hormonally/immunologically driven (Haukisalmi and Henttonen, 2000). However, Kisielewska (1970 b) reported a male bias for this species in bank voles from various localities in Poland with ' excessive infection ', but not those with ' standard' infections. In the Nearctic, Bangs (1985) also reported a male bias for this species in Myodes rutilus whilst Torres et al. (2001) found no sex bias in Eurasian badgers (Meles meles). In fact, little is known in general about the host-parasite relationship of M. muris, apart from reports of its occurrence and those describing its life cycle (Quentin, 1970 ;Vargas, 1977, 1978). The higher prevalence and abundance in female animals is unusual for helminths, since where host sexes differ in susceptibility and resistance, it is more usual to find both prevalence and abundance higher among males, which generally show weaker immune responses to infection and being more vagile are more likely to be exposed to infective stages of parasites (Alexander and Stimson, 1988 ;Poulin, 1996 ;Zuk and McKean, 1996 ;Roberts et al. 1996). One possibility is that female bank voles in our study fed more often on the insect intermediate hosts of this species (Dyer and Olsen, 1967 ;Quentin, 1970 ;Campos and Vargas, 1977) and hence were more exposed to the infection, rather than being less resistant to M. muris. Even fleas may be intermediate hosts (Beaucournu and Chabaud, 1963 ;Miyata, 1939 ;Dyer and Olsen, 1967), although Shogaki et al. (1972) failed to find any evidence of this among the fleas they examined. If the vole fleas can act as an intermediate host infection presumably takes place during the microphagous larval stage, which is likely to be aggregated in vole nests. Pockets of local transmission in and around nests may also be mediated by the other ground insects known to act as hosts for this species. Although males generally carry more fleas (Bajer, unpublished observations), they spend less time in nests, since they have larger territories to defend in the breeding season (Gipps, 1985 ;Ylonen and Viitala, 1991). Clearly M. muris accumulates with host age, a robust finding that was evident in both years of our study through the highly significant effects of age and sex on both prevalence and abundance. This is a reflection of its slow rate of maturation with a 4-6 week pre-patent period (Dyer and Olsen, 1967 ;Quentin, 1970) and probable longevity, as in other closely related spirurids (Quentin, 1969 ;Lowrie, 2003), and is consistent with the idea that mature, rather than juvenile, females are especially at risk of infection, possibly whilst nursing young.
Another species that retained its relative ranking between the three sites was A. tetraptera. In both surveys prevalence and abundance were highest at Pilchy, intermediate at Tałty and lowest at Urwitałt, but despite this apparent stability, there were interesting changes between the surveys. Thus prevalence increased in 2002 at both Urwitałt and Tałty but remained high and constant at Pilchy, whilst abundance dropped at Pilchy and Tałty, and increased at Urwitałt. In this case therefore, although the differences between sites retained their relative ranking, the magnitude and direction of changes with time were not consistent. The increased prevalence combined with reduced abundance is unusual, since these two measures (abundance and prevalence) are more often positively correlated (Shaw and Dobson, 1995 ;Morand and Guegan, 2000). In this case the reduced abundance at two sites in 2002 is attributable primarily to fewer high-intensity infections and hence lower aggregation in 2002 compared with 1999. In a wider context, since the mid 1990s this nematode has shown a long-term trend of increasing prevalence and abundance from year to year in the bank vole populations, that has not yet peaked (Bajer et al. unpublished observations).
C. henttoneni, the only cestode to occur frequently enough to be amenable to quantitative analysis, was more often encountered at Urwitałt in both years and least so at Pilchy. Again, the relative ranking of sites was identical in both years of the survey, although there was an increase in prevalence and abundance in 2002 at all three sites. Clearly, the site effect was the most important factor influencing C. henttoneni.
In previous studies (Behnke et al. 2001) and in other rodents (Montgomery and Montgomery, 1989 ;Behnke et al. 1999Behnke et al. , 2001Behnke et al. , 2004, mean species richness and Brillouin's ID have both increased significantly with host age, and for both measures age explained most of the deviance in statistical models (Montgomery and Montgomery, 1989). Species richness and diversity increase with age because with increasing time after birth there is an increasing cumulative probability for exposure to parasite transmission stages, and, as long as the parasites are long-lived and do not stimulate host protective immunity, their burdens should increase with age . Not surprisingly, therefore, and in common with several earlier studies, the agedependence of mean species richness and Brillouin's ID were both robust findings (Montgomery and Montgomery, 1989 ;Behnke et al. 2001 ;Bajer et al. 2005).
Averaged across all three sites, neither species richness nor Brillouin's ID changed with time (see also Bajer et al. 2005). Nevertheless, the overall means disguised the fact that there were timedependent changes in all sites, but in opposite directions. The rankings of both indices changed completely between the two years (the significant interactions between site and year). Given that most of the common species discussed above maintained their relative ranking across the three sites, and that exclusion of the rare species made no difference at all to the ranking, this is perhaps at first surprising. However, this is clearly the consequence of differences between the sites in the degree to which the prevalence and abundance of the individual common species changed with time. At Urwitałt, the number of recorded helminth species increased from 9 to 12, some of which affected sizeable proportions of the population (e.g. 12 % prevalence for T. martis), and 3 of the common species showed higher prevalence in 2002 (A. tetraptera, M. muris and C. henttoneni). Cumulatively these changes were sufficient to make species diversity and Brillouin's ID highest at Urwitałt in 2002. In contrast Pilchy, which had the highest mean species richness and Brillouin's ID in 1999, had the lowest in 2002. The range of species recorded fell by only 1 but the greatest impact on both measures was from the 51 % decline in the prevalence of H. glareoli by 2002. The prevalence of all other species changed only by less than 5 %.
As discussed earlier (Bajer et al. 2005), the key players in determining infracommunity structure are the common species (often referred to controversially as core species, see Bush et al. 1997). While quantified measures for these species vary moderately from year to year, they nevertheless maintain a relatively stable infracommunity structure in the medium term. In the present study this was reflected in the magnitude of the site effect relative to other factors, and the maintenance of the relative ranking of sites across the two surveys for all of these species. In fact even when there was a significant interaction between site and year (e.g. H. mixtum and M. muris), it reflected either an increased divergence (M. muris) or a reduced difference (H. mixtum) between sites, not a change in relative ranking. The rarer species in the study contributed little to mean species richness or Brillouin's ID because of their low prevalence, other than on the infrequent occasions when prevalence was exceptionally higher (e.g. T. martis at Urwitałt). The contribution of these species could not be analysed quantitatively because infections were so infrequent, hence, in contrast to their contribution to component community structure where they do exert a more important influence on derived summary statistics, their contribution to infracommunity structure was negligible (see Behnke et al. 2008 for discussion of these species).
In conclusion, this study has shown that the helminth infracommunities of bank voles have a rather stable core of species that does not vary much from year to year. The exact location from which voles were sampled was critical, since most of the measures of infracommunity structure varied between sites, even when these were in relative proximity (10-25 km in our study ; Montgomery and Montgomery, 1990). Over the medium term, as represented by a 3-year interval between surveys, most of the indicators of infracommunity structure relating to common species showed the same relative ranking across sites, suggesting a substantial element of predictability and stability, with the key exception of S. petrusewiczi. The picture was more dynamic also at the level of the derived data for infracommunities (mean species richness and Brillouin's index of diversity) and primarily because of the combination of relatively small but contrasting changes in prevalence and abundance of the common, rather than rare, species between the years of the study. It will be interesting to re-assess these sites again in the future, and to identify any longer term influences on the measures quantified here, especially in the context of changing climatic conditions which are causing concern with respect to the spread of existing parasitic infections and emerging infectious