Consistencies in post-dispersal seed predation of temperate fleshy-fruited species among seasons, years and sites



1. Seed predation of 12 fleshy-fruited species was recorded in experimental dishes under early successional forest in south-west Germany (four seasons 1992 and 1993) and in south England (summer 1995). On each occasion, 200 dishes were laid out, each containing five seeds of a given species. The mean time for three or more seeds to be removed was taken as a measure of granivore preferences. We tested correlations of these preferences with several physical and nutritional seed traits.

2. Live trapping and selectively accessible dishes indicated that rodents were the major granivores (Apodemus sylvaticus, Apodemus flavicollis, Clethrionomys glareolus); no predation by birds or insects was observed. The rank order of the rodents’ seed preferences was consistent among sites, seasons and years, but mean predation differed between species, sites and seasons. Seed predation was highest in summer and lowest in winter. Possible implications of the seasonal pattern in predation risk are discussed.

3. The preferences of rodents were significantly correlated with the species-specific viability of seeds (i.e. percentage of sound seed) in five of eight experiments and with the percentage of water in the embryo-plus-endosperm fraction (EEF) in four of eight experiments. Surprisingly, preferences were not correlated to seed mass, EEF mass or nitrogen concentration. Predation was lowest among toxic species (Berberis vulgaris, Euonymus europaeus, Sambucus nigra) and among species with woody endocarps (Cornus sanguinea, Crataegus spp.).


Small mammals often play a crucial role during plant succession; for example, from semi-desert to grassland (Brown & Heske 1990) or from open grassland to forest (Jensen & Nielsen 1986). Although the effect of seed predation on succession has been discussed for more than two decades (Janzen 1971), it is still controversial (Davidson 1993; Hulme 1996a). Rodents have been shown both to enhance and to hinder succession. Their habit of collecting seeds and then burying them elsewhere, as distant as 20–30 m from the collecting site, aids the dispersal of the plants (Abbott & Quink 1970). Although the seeds are clearly stored for future consumption, seeds frequently survive to germination (Vander Wall 1990). On the other hand, seed predation by rodents, particularly after dispersal, is one of the most important mechanisms of slowing down old-field succession (De Steven 1991; Gill & Marks 1991). It follows that any seed preference of the granivores has a potential effect on species-specific regeneration rates and on colonization, and hence might affect the community structure.

Although many studies have focused on seed preferences of small mammals in deserts (Hay & Fuller 1981; Abramsky 1983; Kelrick et al. 1986) and in temperate regions of North America (Mittelbach & Gross 1984; Whelan et al. 1990; Willson & Whelan 1990; Myster & Pickett 1993; Schwantes Boman & Casper 1995), seed predation in European temperate vegetation is poorly investigated (but see Hulme 1996b; Hulme & Borelli, in press). Moreover, we are not aware of any study in which consistencies of seed predation were compared among seasons, years and sites. Because seed rain is often seasonal, for example, in temperate fleshy-fruited species (Kollmann 1994), seasonal differences in seed predation have a high potential impact on recruitment.

The two objectives of this paper are to investigate (1) spatio–temporal consistencies in seed preferences and (2) the reasons underlying these preferences. We tested seed predation of rodents among 12 fleshy-fruited species in early successional forest. Selectivity was compared among four seasons and 2 years in south-west Germany and during one summer in south England. The animals responsible for predation were identified with selectively accessible seed dishes and live traps. Where strong and consistent preferences occurred, we attempted to explain them by looking at basic seed characteristics.

Study sites, materials and methods


This study deals with 11 fleshy-fruited European tall-shrubs (Berberis vulgaris, Cornus sanguinea, Crataegus laevigata, Crataegus monogyna, Euonymus europaeus, Ligustrum vulgare, Prunus spinosa, Rosa canina ssp. dumalis, Rubus fruticosus agg., Sambucus nigra, Viburnum lantana) and the tree Prunus avium. Nomenclature follows Tutin et al. (1964–1980) only generic names are used subsequently except for Crataegus spp. and Prunus spp. The species, from 10 genera within six families, exhibit a range of seed masses and differ greatly in anti-herbivore protection (Table 1). Fruit of these species were collected from three to five individual plants per species in the German study area (Kaiserstuhl) 1–3 years prior to the experiments. The fleshy tissue of the fruit was mechanically removed by sifting under cold running water. Seeds were dried at 30 °C, and thereafter stored at 5 °C.

Table 1.  . Seed characteristics of the fleshy-fruited species used in the predation experiments (n = 30 seeds; means ± SE). Free water of the embryo-cum-endosperm fraction (EEF) is calculated as reduction of fresh mass after 48 h in 70 °C. Information on toxicity is taken from the literature (Paris 1963; Hegnauer 1964; Cooper & Johnson 1984; Lang 1987; Snow & Snow 1988). Ranking of the species follows apparent preferences of rodents Thumbnail image of


The physical characteristics of each species were measured using samples of 30 seeds. The fresh and dry masses (70 °C, 48 h) of the whole seed and those of the embryo-plus-endosperm fraction (EEF) were measured to the nearest 0·1 mg. The testa of Berberis, Euonymus and Ligustrum was too thin for separation from EEF, so the whole-seed mass is given in these cases. The proportion of sound and apparently viable seed (‘seed viability’) was determined by dissecting and visually inspecting the embryos and also by comparing the EEF dry mass with the average; a value less than 33% of the average was defined as ‘dead’.

Because many granivores remove the testa before eating the EEF of seeds, only this fraction was analysed for nitrogen concentration. Dried samples of EEF (100 mg) were subjected to Kjeldahl digestion and analysed for ammonia using a variation of the Bertholet reaction, with salicylate instead of phenol and dichloroisocyanurate to give hypochlorite ions; nitroprusside was added as a catalyst. The reaction was arranged as a continuous flow analysis; the indophenol produced was measured at 650 nm (Method sheet CW2–008–11, Chemlab Instruments Ltd, Hornchurch, Essex, UK). A standard sample of leaf material was included in each batch to test the accuracy of the procedure.


South-west Germany

In south-west Germany, the study was situated under early successional forest, at Haselschacher Buck nature reserve, Kaiserstuhl, a small mountain range in the Upper Rhine valley (48° 06’ N, 7° 40’ E; 470 m a.s.l.; 9·9 °C mean annual temperature, 632 mm annual precipitation; Kollmann 1994). The site was situated on a north-facing slope with an inclination of about 15°; the soil substrate was calcareous loess on volcanic bedrock. The forest stand of about 1 ha consisted mainly of Corylus avellana (4–6 m height; 70–90% cover) with a few interspersed trees (Quercus petraea, Tilia cordata); the herbaceous layer was sparse (Mercurialis perennis, Urtica dioica) and covered less than 5% of the ground. The forest was surrounded by calcareous grassland dominated by Bromus erectus, Brachypodium pinnatum and Salvia pratensis with scattered pioneer scrub; the age of the successional forest was about 50 years.

South England

The second study was also situated under early successional forest, at Box Hill nature reserve near Dorking, Surrey (National Grid reference: TQ178528; 100 m a.s.l.; 750 mm annual precipitation). The site was situated on a north-facing slope on chalk with an inclination of 10°; it was a strip of forest 100-m wide, bordered for 500 m on both sides by chalk grassland. The site was a pasture about 50 years ago, and was then colonized by shrubs and Betula pendula. It is now dominated by Fraxinus excelsior with dead Betula and shrubs in the understorey. The tree layer averaged 10–20 m in height and 50–60% in cover. In the shrub layer C. avellana was most prominent (height 2–6 m, cover 60–70%); the herbaceous layer (0·2–0·4 m) was dominated by M. perennis (about 50% cover).


Seed dishes

Seed predation was studied using green plastic seed dishes (12-cm diameter), with slits for drainage and covered with a transparent PVC screen (15 cm × 15 cm), supported by an 8-cm metal pole. This set-up was identical for both study sites and similar to Abramsky (1983), Willson & Whelan (1990) and Hulme (1996b); for critical comments on the seed dish technique see Kelrick et al. (1986).

In the German site 100 dishes were arranged in regular 10 × 10 grids in a 27 m × 27 m quadrat; each dish contained five seeds of one species and was situated 3 m from its nearest neighbour; so each dish was a pure culture of one species of seed. We experimented with five species (Cornus, Crataegus laevigata, Ligustrum, Prunus avium, Rosa) in three seasons of 1992 (not autumn) and with 10 species (the above list, plus Berberis, Euonymus, Prunus spinosa, Sambucus, Viburnum) in all four seasons of 1993. Two quadrats were used in 1992 and four in 1993; that is 40 dishes per species in both years. The dishes remained in the same position for 2 (1) years, and seeds were added on 11 January, 6 April and 14 July in 1992, and on 26 January, 16 April, 22 July and 22 October in 1993. The disappearance of seeds was checked after 1, 2, 4, 7, 14, 28 and 56 days. Each of the eight experiments started during a period of fair, sunny weather when rodents are particularly active (Gurnell 1975).

In south England 200 dishes were arranged in two 27 m × 27 m quadrats in the summer 1995. The same seed material was used, except that Prunus avium was replaced by Rubus and Crataegus laevigata by the locally common Crataegus monogyna. Within each quadrat 10 dishes were randomly allocated to each species, that is 20 dishes per species. One day after setting the dishes, the seeds were added and then checked daily for the first 10 days and then at longer intervals, after 20, 30 and 50 days. During the experiment at this site (20 June to 8 August) the weather was predominantly dry and sunny.

Dishes with limited access to granivores

The standard seed dishes allowed access to any type of granivores. A series of dishes that restricted access to rodents, birds or insects was set up, thereby checking that rodents were the major granivores. First, a dish was raised on a stick 1 m above the ground, making it only accessible to birds and flying insects. Second, a 1 cm wire mesh on the dish prevented birds and rodents but allowed access to insects. Third, an upside-down plastic dish with sticky tape was added to stop insects; the whole set-up was covered with a 4-cm wire mesh to exclude birds, thereby allowing only access of small mammals. There were also dishes made that had (1) the PVC roof removed, (2) the metal pole removed and (3) the plastic dish replaced by a glass Petri dish. These dishes were used to test whether seed removal was affected by the design of the seed dish. Five replicates of each dish type were baited with five seeds of Prunus avium, an attractive seed to granivores (Kollmann 1994), and were randomly ordered in transects near the experimental plots at both sites.


The population of small mammals was sampled using Longworth live traps (Penlon, Abingdon, UK). At the English site, 36 traps were set out on 22 June 1995; they were installed in pairs so that a catch would not prevent another and the pairs were regularly arranged within the two experimental quadrats. After allowing the rodents to get accustomed to the traps, they were first baited on 25 June and then checked morning and evening until 29 June. In Germany six Longworth traps were set out from April to September 1993 (13 nights) in six patches of mature scrub adjacent to the study site (cf. Kollmann 1995). In both Germany and England, the traps were baited with Prunus avium and any catch was identified and labelled before being released. In traps that had been used the hay bedding was changed, because residual odour affects subsequent catches (Stoddart & Smith 1986; Gurnell & Little 1992).


The rate of disappearance of seeds was investigated using ‘survival analysis’. A dish was considered ‘empty’ when two or less seeds remained; this definition allowed both for accidental losses of one to two seeds and for persistence of sterile seeds. The data were analysed using a ‘censoring’ method, because observations did not continue until all dishes had been emptied. If the rate of removal of seeds is related only to the number of dishes that contain seeds at a given time, one would expect the number of extant dishes to decline exponentially. The Weibull function is a more general means of describing survivorship curves and takes the form of a negative exponential when the parameter ‘α’ is set to unity but has a steeper decline when α is less than one. We used a macro in the GLIM statistical package (Glim 4·0, Royal Statistical Society, London, UK) to fit Weibull functions to each curve, and tested whether the curves differed from exponential using χ2 tests (Crawley 1993; pp. 314–331).

Once the values of α had been established by the Weibull macro, significant differences between species, seasons, sites and years were tested by the method of model simplification (Crawley 1993). The deviance of terms removed from a model were tested using a χ2 distribution. We calculated the time-to-disappearance of each species (at each season, site and year) by back-transforming estimate from the GLIM models.



The results from the limited-access trays suggest that small mammals were responsible for seed removal, because no seeds were eaten in the dishes accessible only to birds or insects, whereas all seeds disappeared from dishes accessible to small mammals. In both study sites the presence of urine, droppings and seed remnants in the dishes indicated that rodents were the major granivores. The rodents’ feeding was not affected by the nature of the dish. The live trappings confirmed the presence of rodents: Apodemus sylvaticus L. and Clethrionomys glareolus Schreber were trapped in both the German and the English site, while Apodemus flavicollis Melch. was trapped only in Germany. The average density of rodents was far greater in Germany than in England (0·58 vs 0·10 rodents trap–1 night–1).


Survivorship of seeds over time

The survivorship curves were clearly different among the 12 fleshy-fruited species (Fig. 1). The rate of seed removal was greatest for Prunus spp. and Viburnum, intermediate for Rosa, Ligustrum, Rubus and Cornus, and rather low for Berberis, Sambucus, Euonymus and Crataegus spp. Because the y-axes of Fig. 1 are logarithmic, a constant probability of predation would result in a linear decrease with time. In fact some curves are steeper at first and then flatten off, suggesting that predation was greatest in the first few days after exposure. The fitting of Weibull functions confirmed that the probability of predation was not constant in the cases of Cornus, Crataegus spp., Ligustrum and Rosa (α = 0·85), although the other species had curves that were not significantly different from log-linear.

Figure 1.

. Survivorship curves for each species (averages across all dates and sites). The order of species follows mean time-to-disappearance of seeds.

Spatio–temporal consistencies of seed predation

Much of the deviance in the time-to-disappearance of seeds was explained by differences between species (r2 = 0·20–0·39, P < 0·0001; Table 2) and by differences between countries (r2 = 0·17–0·36, P < 0·0001). The rates of predation in winter 1992 and 1993 were very similar (r2 = 0·00, P > 0·50), and although there were significant differences between the rates of predation in spring 1992 vs 1993 (r2 = 0·02, P = 0·001), and in summer 1992 vs 1993 (r2 = 0·014, P < 0·001), the year effects only explained a relatively small amount of the total deviance. The rate of predation was ranked summer > spring > autumn > winter (Fig. 2).

Table 2.  . Analysis of deviance of the differences in seed predation among species, years and sites. Mean time-to-disappearance in the removal experiments is taken as a measure of predation; Crataegus monogyna and Rubus had to be excluded because they were tested only in the English site. ****P < 0·0001, ***P < 0·001 Thumbnail image of
Figure 2.

. Differences in seed predation among seasons and years in the German (D) and the English site (UK). Mean time-to-disappearance serves as a measure for predation (median with 25–27% percentile).

The preferences of rodents were notably consistent among seasons and years. Significant rank correlations were found for most of the pair-wise comparisons of the times-to-disappearance at the German site but there were few significant correlations between the German and the English site (Table 3).

Table 3.  . Consistencies of rank order of species among seasons (wi., winter; sp., spring; su., summer; au., autumn), years (1992, 1993) and sites, tested with Pearson product moment correlations on log mean time-to-disappearance. ***P < 0·001, **P < 0·01, *P < 0·05; not significant where no indication Thumbnail image of


We sought to explain the preferences of rodents in terms of seed characteristics (cf. Table 1). By far the most consistent correlation was found between seed viability (i.e. percentage of sound seed) and the logarithmic time-to-disappearance; there was a significant Pearson product moment correlation in five of the eight experiments (P < 0·05). The percentage water in the endosperm-plus-embryo fraction (EEF) was significantly correlated with logarithmic time-to-disappearance in four of the eight experiments (P < 0·05). We had expected the preferences to be related to seed size (cf. Hulme 1993), (percentage) EEF mass and nitrogen concentration, but no such correlations were found. It is possible that the effects of these factors were obscured by the influence of differences in seed viability. When species with low seed viability (< 85%) were excluded from the analysis seed mass was negatively related to time-to-disappearance (P = 0·03–0·15, four seasons 1993) but this result was strongly influenced by the two largest seeds that were quickly eaten (Prunus avium, Prunus spinosa).



Rodents, the primary granivores present, demonstrated species-specific selectivity among the 12 fleshy-fruited species and the seed preferences were consistent among seasons, years and partly among sites, even though the rate at which seeds were eaten varied between seasons. Of course, our experimental design (one species of seed per dish) did not allow us to test for real preferences because the small mammals were not confronted with a choice of seed per dish. However, as indicated below by the observed mobility of the rodents they probably encountered more than one dish per site and night, that is dishes with different seed species. Thus, we investigated seed preferences sensu lato.

These preferences probably reflect the undisturbed behaviour of the rodents, because the rates of predation were correlated with the percentage of destroyed seeds in soil samples from the German site (Kollmann 1994). Moreover, a similar rank order of species was found under direct observation of A. sylvaticus in the laboratory (S. M. White, unpublished results). The literature does not provide much information for the comparison of seed preferences. However, Jensen (1993) reported a similar rank order for A. sylvaticus feeding on Rubus, Rosa, Sambucus and Crataegus monogyna; only Prunus spinosa was less attractive than in our study. Although the rodents observed are partly cache hoarders, most seeds were probably consumed immediately because the dishes were often littered with seed remnants after visitation by the granivores.

The fact that only a few correlations were found between preferences of the rodents and the seed traits investigated suggests that other factors besides seed viability must be important. Other studies have reported a correlation between predation and energy content of the edible part of a seed (Reichman 1977; Vickery 1984; Kerley & Erasmus 1991). However, EEF mass showed no significant correlation with the preferences and EEF is strongly correlated with the energy content of seeds (M. Fenner, personal communication). Rodents often prefer to consume seeds that are rich in lipids (Kerley & Erasmus 1991) but unfortunately this trait has not been measured.

Toxicity is most probably a major determinant of feeding preferences (Janzen 1969; Hulme 1993). For example, Euonymus, Berberis, Prunus spp. and Viburnum all produce large seeds with relatively high nutrient contents and high viability which suggests that they would be susceptible to granivory. In fact Euonymus and Berberis are known to contain a host of toxic substances and are not palatable to rodents (Table 1); we observed that mice often started to eat these seeds, getting as far as removing the testa, but then abandoning them (Kollmann 1994). On the other hand Viburnum, with no known toxins, and Prunus spp., with cyanogenic glycosides (not thought to be toxic to rodents, Cooper & Johnson 1984), were readily eaten.

Because the information on seed traits is not sufficient to explain the strong preferences of rodents, they cannot be discussed on the basis of optimal foraging (Pyke, Pulliam & Charnov 1977; Reichman 1977; Vickery 1984). For this objective, feeding trials under controlled conditions (cf. Price 1983) are required to measure the harvesting and handling times, and to see whether differences in abundance of seeds affect the rodents’ choice. In addition, a thorough nutritional and toxicological analysis of the seeds is necessary in order to determine the benefits gained by the rodent per seed consumed.


The live trapping suggested that the resident population of the English site in summer 1995 was smaller than in the German site in summer 1993 and this correlates with the lower seed predation in the English site (Fig. 2). It also showed that rodents moved up to 50 m distance between quadrats in the English site. Such movements are not uncommon (Kikkawa 1964) and suggest that an individual should have had no difficulty in encountering most of the experimental dishes containing different species during one foraging trip. Finally, the trapping revealed that A. flavicollis was probably not present in the English site. Because A. flavicollis is larger than A. sylvaticus, and can cope with bigger seeds (Flowerdew 1993), this might contribute to the site-specific rank order of the species. Because successional changes in vegetation structure can also affect the order of preferences (Myster & Pickett 1993), the vegetation, which did differ in some respects between the two sites, might be another factor for differences in predation.

The different intensities of predation among seasons (summer > autumn > spring > winter) may also reflect temporal shifts in the population density of rodents. In the temperate zone rodent populations often undergo seasonal changes owing to their reproduction cycle, with high densities in summer and autumn (Halle 1993; Jedrzejewski & Jedrzejewski 1996). Highest predation coincided with the annual peak in seed rain of fleshy-fruited species in the study sites (autumn > summer > winter > spring; Kollmann 1994). However, the seasonal pattern in seed predation probably does not reflect dietary changes, because seeds appear to be most prominent in the diet of rodents during the winter months, whereas insects are more important in late spring and summer (Watts 1968; Halle 1993).


The observed seasonal differences in seed predation have potential effects on the regeneration of fleshy-fruited species (cf. Hulme 1996b) and add a temporal component to the spatial ‘regeneration window’ of these species in a successional sere (cf. Kollmann 1995). Species fruiting in summer and autumn have the disadvantage of seed dispersal during the period of most intensive predation. Additionally, the probability of seed survival depends on the time interval between dispersal and germination, although the results of this study indicate that the predation risk for some species is highest in the first weeks after dispersal when the seed odour is still fresh (cf. Price & Jenkins 1986). In central Europe seed rain of fleshy-fruited species is highest in late summer and autumn, whereas germination mostly occurs in spring (April to June, Kollmann 1994). Therefore, species fruiting in summer (Prunus avium, Viburnum) are exposed to post-dispersal seed predation for a much longer period than those ripening in late winter and early spring (Hedera helix, Ligustrum). So far no investigations have focused on the long-term effects of seasonal predation in the context of temporal patterns of seed rain and germination.

The functional effect of the species-specific differences in seed predation remains also to be investigated. Significant consequences for recruitment and colonization of new sites are to be expected, although preliminary results indicate that the ability to establish a seed bank and differences in germination and establishment may be of considerable importance (Kollmann 1994; Kollmann & Reiner 1996; Grubb et al. 1996). Spatio–temporal conflicts between subsequent stages in the life cycle of plants are well known (Schupp 1995; Kollmann & Schill 1996) and seed predation may be insignificant where other stages of recruitment are limiting (Andersen 1989; Hulme 1996b). However, these questions can only be answered when experiments are set up that exclude rodents over several years (e.g. Heske, Brown & Guo 1993).

J. Kollmann et al.

J. Kollmann et al.

J. Kollmann et al.


We thank Sophia Kollmann and Lucy Muir for help with the fieldwork, and Glyn D. Jones for technical assistance with seed analysis. Phil E. Hulme improved the manuscript greatly with critical comments. The research of J.K. was supported by grants from the Landesanstalt für Umweltschutz, Karlsruhe (PAÖ-No. 209118·01), and the Reinhold-Tüxen-Gesellschaft, Hannover.