State-dependent invasion windows for prey in size-structured predator–prey systems: whole lake experiments



    1. Department of Ecology and Environmental Science, Umeå University, SE-90187 Umeå, Sweden; and *Institute for Biodiversity and Ecosystem Dynamics, University of Amsterdam, POB 94084, NL-1090 GB Amsterdam, the Netherlands
    Search for more papers by this author

    1. Department of Ecology and Environmental Science, Umeå University, SE-90187 Umeå, Sweden; and *Institute for Biodiversity and Ecosystem Dynamics, University of Amsterdam, POB 94084, NL-1090 GB Amsterdam, the Netherlands
    Search for more papers by this author
  • and * PÄR BYSTRÖM

    1. Department of Ecology and Environmental Science, Umeå University, SE-90187 Umeå, Sweden; and *Institute for Biodiversity and Ecosystem Dynamics, University of Amsterdam, POB 94084, NL-1090 GB Amsterdam, the Netherlands
    Search for more papers by this author

L. Persson, Department of Ecology and Environmental Science, Umeå University, SE-90187 Umeå, Sweden. Tel.: +46 90 7866316. Fax: +46 90 7866705. E-mail:


  • 1In size-structured communities where individuals grow in size over their life cycle, interactions between species will shift between competitive and predatory interactions depending on size relationships. The outcome of interactions will subsequently depend on the strength of competitive and predatory interactions, respectively.
  • 2In a whole lake experiment including four experimental lakes, it was tested under which conditions the competing prey, roach Rutilus rutilus, could successfully recruit into systems previously occupied by the predator, perch Perca fluviatilis. Two replicated introduction experiments were carried out 3 years apart.
  • 3Roach were able to successfully recruit into three of the four experimental lakes of which two were also inhabited by the top predator pike Esox lucius. Resource levels were unrelated to whether roach could successfully recruit into the systems as recruiting roach in all years were feeding close to their maximum rate.
  • 4High population fecundity of roach and low predation pressure by perch combined were necessary ingredients for successful recruitment and the presence of only one of these conditions did not result in successful recruitment.
  • 5It is hypothesized that, although roach were able to successfully recruit into one lake with only perch present in addition to the two lakes that also inhabited pike, long-term coexistence of roach and perch depends on the presence of another top predator (e.g. pike) selectively preying on perch. This hypothesis was supported by data on co-occurrence of perch and roach in different lakes.
  • 6Overall, the results are in accordance with expectation of size-structured life-history omnivory theory suggesting that coexistence between top predator and intermediate consumer is fragile.


Body size is undoubtedly the most important trait of an individual, affecting its foraging rate, metabolic demands, prey selection, risk of being eaten and fecundity (Peters 1983; Werner 1988; Kooijman 2000; Cohen, Jonsson & Carpenter 2003; Brown et al. 2004). Moreover, the circumstance that many predators grow over a substantial part of their life cycle means that both the intensity and nature (competition or predation) of interactions will change over the life cycle of the predator (Werner & Gilliam 1984; Persson 1988; Wilbur 1988; Olson, Mittelbach & Osenberg 1995; Byström, Persson & Wahlström 1998). At small sizes, the predator may compete with its prey whereby the prey may limit the extent to which predator individuals are recruited to larger predatory stages, inducing an interspecific competitive juvenile bottleneck in the predator (Neill 1975; Lasenby, Northcote & Fürst 1986; Persson 1988; Olson et al. 1995). On the other hand, prey species may have a positive effect on the performance of larger stages of the predator (Persson 1988; Olson et al. 1995).

Studies on size-dependent competitive and predatory interactions have largely focused on changes in relative abundance of predator and competing prey with environmental conditions and less on the possibility for predator and competing to coexist at all (Persson 1988; Jeppesen et al. 1997; Haertel, Baade & Eckmann 2002). Neill (1975) showed experimentally that size-dependent competitive interactions between small and large cladoceran zooplankton lead to the extinction of the larger species through an interspecifically induced competitive juvenile bottleneck in the larger species. However, in this experiment no predatory interaction was present. Theoretically, coexistence between top predator and competing prey has been addressed in intraguild predation (IGP) theory where interactions between size-structured predator and prey represent a special form of IGP termed life-history omnivory (Polis et al. 1996). Recent theoretical investigations of the life-history omnivory system show that the presence of a food-dependent development rate in predators and competing prey has strong effects on the possibility for coexistence between predator and competing prey (Van der Wolfshaar, De Roos & Persson 2006). Compared with nonstructured omnivorous models, size-structured omnivory with food-dependent development dramatically decreases the opportunity for coexistence between predator and prey (Diehl & Feissel 2000; Mylius et al. 2001; vs. van der Wolfshaar et al. 2006). An important reason for this outcome is that food-dependent growth in the predator causes a positive feedback between resource levels and predator attack rate on prey. This occurs because increased resource levels promote the growth of young predators increasing their per capita predation efficiency on the prey later in life (van der Wolfshaar et al. 2006).

While size-structured life-history omnivory theory thus predicts a decreased likelihood for coexistence between predator and competing prey, at intermediate productivity levels alternative stable states with either prey and resource or predator and resource are likely to occur (van der Wolfshaar et al. 2006). The success of predator and prey, respectively, is hence dependent on the initial state of the system. In a previous whole-lake experimental study it has been shown that the introduction of competing prey roach Rutilus rutilus L. into systems inhabited by the predator perch Perca fluviatilis L. led to a reduction in the shared zooplankton resource and thereby decreased individual growth and increased mortality in perch (Byström et al. 1998; Persson et al. 1999). Here it is first addressed whether this competitive effect of the introduced roach on the predator perch depends on the state of the perch population given that a competitive effect of roach on perch is a function of individual perch size (Byström et al. 1998). This was done by comparing the response of the perch population between two roach introduction events with different perch population size distributions at the date of the roach introduction. Second, it was addressed under which conditions the introduction of roach also resulted in a positive population growth rate of roach. Finally the question was addressed whether a measured positive population growth rate of roach is only a transient phenomenon and long-term exclusion of roach is likely. The latter question relates to the theoretical expectations from size-structured theory that food-dependent growth in predator and prey demotes coexistence (van der Wolfshaar et al. 2006).

Three main factors were tested in relation to roach performance. First it was tested whether successful roach recruitment measured as successful survival of roach to an age of 1 year old depended on roach reproductive output. Second, it was tested whether successful roach recruitment depended on perch predation pressure. Finally, it was tested whether resource limitation could explain roach recruitment variation. Variation in perch predation pressure could potentially come from three sources: intrinsically driven fluctuations in perch (Persson et al. 2003), roach-induced increased mortality in perch (Byström et al. 1998; Persson et al. 1999) or overall variation in predation pressure among experimental lakes due to the presence of other predators (Persson et al. 1996). The latter resulted from that 2 of the 4 experimental lakes investigated inhabited a predator on perch, the northern pike Esox lucius L.

Materials and methods

species configuration studied

Roach and perch are two commonly occurring species in Eurasian systems that dominate the lake fish community of many north European lakes. Perch (top predator) is an ontogenetic omnivore starting out feeding on zooplankton when small, to thereafter shift to macroinvertebrates, to finally become piscivorous (Persson 1988; Hjelm, Persson & Christensen 2000). Roach (intermediate consumer) also starts to feed on zooplankton to shift to a diet consisting of a mixture of macroinvertebrates and plant material when increasing in size (Hjelm & Persson 2001). Roach is a superior competitor to perch on the shared resource zooplankton in terms of a higher attack rate and a lower critical resource demand (resource level needed for maintenance on the shared resource zooplankton) (Byström & Gàrcia-Berthóu 1999). Numerous experiments from enclosures to whole lakes have also demonstrated the competitive superiority of roach (Persson 1987; Persson & Greenberg 1990; Byström et al. 1998).

The top predator pike is also very common in Eurasian systems and preys on both perch and roach. Related to a larger gape size, pike can take larger prey fish than perch while pike also exhibits a selectivity for perch over roach due to the latter's higher evading capacity (Persson et al. 1996; Nilsson & Brönmark 2000). Owing to its gape morphology, pike hardly feeds at all on zooplankton even when small leading to that pike is much less susceptible to competing prey than perch.

lakes descriptions and roach introduction experiments

The study was carried out in four small (2·4–9·3 ha) adjacent unproductive lakes situated in middle Sweden (64°29′ N, 19°26′ E) (Persson et al. 1996). The lakes are situated in a sandy area with slow-growing pine Pinus silvestris forests, reindeer lichens Cladonia rangiferina and lingon berries Vaccinium Vitis-idaéa as ground cover. The lakes have a maximum depth between 9 and 14 m, but due to a more extended area of shallow water, two of the lakes (Lakes Abborrtjärn 1 and 2, LAT 1 and 2 in the following) have a lower mean depth than the other two lakes (Lakes Abborrtjärn 3 and 4, LAT 3 and 4 in the following) (see also Persson et al. 1996 and Wahlström et al. 2000).

Before the introduction of roach, all lakes were inhabited by perch and two of them (LAT 1 and 2) also had small populations of pike leading to a lower perch density in these lakes (Persson et al. 1996) (Fig. 1, year 1992). No other fish species are present in the lakes. Roach were introduced to one of the lakes previously inhabited by perch only (LAT 4) and to one of the lakes previously inhabited by both perch and pike (LAT 2) in the spring and late autumn of 1993 (Table 1). The spring stocking consisted of large roach (146·7 ± 53·1 mm, mean ± 1 SD) and the autumn stocking of small roach (63·3 ± 5·5 mm, mean ± 1 SD). The size structures and numbers of the introduced roach corresponded to natural densities found in lakes of similar productivity and species composition (including perch and pike). In autumn 1996 and spring 1997, roach were also introduced into LAT 1 and 3 at numbers and size distributions similar to that stocked in LAT 2 and 4 in 1993 (Table 1). Effectively, stocking with two size groups of roach meant that the population fecundity was expected to increase substantially when the small size category matured (on average 3 years after stocking) due to the higher numbers at which they were stocked (Table 1). In a conservative sense, the extent of replication in the whole lake experiments was for logistic reasons small (2*2). However, it should be kept in mind that as large adult roach survived and reproduced every year in all lakes from the start of their introduction up to the end of the study, each individual lake has been subjected to repeated (6–11) yearly recruitment attempts by roach over the study period (34 recruitment attempts in total).

Figure 1.

Population numbers (mean ± 95% CL) of perch ≥ 2 years in (a) LAT 1 and 2 and (b) LAT 3 and 4 from 1992 to 2004. Arrows at top show the years for stocking of roach in LAT 2 and 4 (autumn 1993) and LAT 1 and 3 (autumn 1996 and spring 1997).

Table 1.  Numbers of roach (no. per ha) added to Lakes Abborrtjärn 2 and 4 in 1993 and to Lakes 1 and 3 in 1996/1997
Roach 50–100 mm (autumn 1993)196238
Roach > 100 mm (spring 1993) 71 84
Roach 50–100 mm (autumn 1996)207214
Roach > 100 mm (spring 1997) 85 82

sampling methods

Perch 1 year old and older

Sampling methods for perch 1 year old and older were based on capture-recapture methods. Methods for capturing perch (≥ 2 years) including traps and fyke nets have been described in previous papers (cf. Persson et al. 1996, 1999). One-year-old (OYO) perch were electrofished in spring from a boat when they were concentrated along the shore. In years with too low numbers of OYO perch to allow population estimates based on mark–recapture, population abundance was calculated based on a regression of numbers estimated by mark–recapture on numbers captured during the standardized spring electrofishing (Byström et al. 2003). OYO perch numbers were also estimated based on total trap and fyke net captures in spring each year. All captured perch were measured (to nearest millimetres) and weighed (to nearest 0·1 g).

Roach 1 year old and older

For large roach, reliable estimates of population size based on mark-recapture have only been successfully obtained in one lake (LAT 4) (Persson et al. 1999). Therefore fyke net and trap catches in spring and gill net catches in August were used as measures of the relative abundances of roach. Gill net catches were obtained with gill nets (benthic and pelagic) of survey link type (see Persson et al. 1999). The capture in the benthic and pelagic net was weighed for the volume of the lake section that the respective net represented to arrive at a habitat volume corrected capture per net (catch per unit effort, CPUE). All captured roach were measured (to nearest millimetre) and weighed (to nearest 0·1 g). For the two lakes where roach stocking took place in 1993, gill net catches in LAT 4 were higher than those in LAT 2 despite that snorkelling data suggested similar densities. These higher gill net catches in LAT 4 can be related to the considerably smaller surface area of the former leading to an expected overestimate of roach density in LAT 4 relative to LAT 2 (see Persson et al. 1999). In the analysis of recruitment success, this bias was not corrected for, which means that found differences in recruitment success between treatments (stocking year) are conservative (not correcting leads to higher within treatment variance).

Young-of-the year perch and roach

Larval perch and roach were sampled quantitatively once a week during 5 weeks following hatching in early (perch larvae) and middle (roach larvae) of June. The sampling was done with a Bongo-trawl attached to a small boat. A more detailed description of the equipment and methods used is given in Byström et al. (1998).

Recruitment success of roach was measured as the number of roach in the size range of 40–100 mm (mainly 1-year-old and 2-year-old roach) captured in traps and fyke nets in spring every year. It should be noted that roach captured in this size range up to at most 3 years after stocking may have originated from the small-sized roach stocked (Table 1). Thereafter, capture of roach in the size range 40–100 mm reflects successful recruitment of hatched roach to an age of at least 1 year old. We focused on this size range for assessing successful roach recruitment, because size-dependent life-history omnivory theory predicts that extinction of roach result from perch predation on YOY roach (van der Wolfshaar et al. 2006) and the optimal prey size–predator size ratio of perch is as small as 0·15 (Persson et al. 2004).

Samplings in LAT 1 and 2 were temporarily stopped in 2003, hence analyses including these two lakes do not go beyond this year. As samplings were resumed in LAT 1 and 2 in 2005, data are presented on roach numbers and population size structures from spring (trap and fyke net catches) and August (survey gill nets) of 2005 to provide the most recent data on roach performance in all four lakes.

zooplankton resources and estimates of resource limitation

Zooplankton were sampled eight times during May–October in every year. Samples were taken at three pelagic stations in each lake and were collected with a 100 m-mesh net (diameter 25 cm) drawn vertically at an approximate speed of 0·5 m s−1. One tow was made at each pelagic station from the thermocline (estimated with a thermistor) to the surface. Zooplankton samples were preserved with Lugol's solution. In the laboratory, animals were classified by species, counted and the lengths of 15 individuals (all, if fewer were collected) of each species from each sample were measured in an inverted microscope. Lengths were transformed to biomass using regressions relating length to dry weight (Bottrell et al. 1976).

For the present paper, zooplankton data from the period end of June to the beginning of September (average based on four samples during this period) was used in the analysis of the degree of resource limitation in different years. This period has been shown to be the period when zooplankton resources are depressed to the largest extent and hence the likelihood for resource limitation is highest (Byström et al. 1998; Persson et al. 2004). Consumption rates of YOY roach was subsequently calculated and compared with maximum consumption rates. Resource limitation was expressed as the ratio of estimated consumption/maximum consumption rate (no resource limitation at ratio 1). Consumption rates were calculated from size-dependent functional response curves based on a 0·5 mm cladoceran prey (see Hjelm & Persson 2001). Functional response parameters are only accessible for cladocerans, hence the estimates of consumption rates relative to maximum consumption rates do not include other prey types (copepods and macroinvertebrates) eaten by roach. Still, cladoceran zooplankton constitutes the main proportion (75%) of the diet of YOY roach (Persson et al. 2000).

predator attack rate measures

To estimate the predation pressure by perch on YOY roach in the different lakes in different years, a total size-dependent population attack rate of perch was calculated based on the estimated number and size distribution of perch 1 year old and older in May–June each year. The equation for the size-dependent piscivorous attack rate of perch is given in Persson et al. (2004) where also parameter values used can be found. The calculations were based on a 17 mm roach as prey that represents the average size of YOY roach when the main predation mortality induced by perch occurs (see Results below).


development of perch≥ 2 years in relation to roach introductions

Before the first invasion experiment with roach in 1993, the population sizes of perch ≥ 2 years were lower in LAT 1 and 2 than in LAT 3 and 4 related to the presence of pike in the two former lakes (Fig. 1) (Persson et al. 1996). In the summer of 1994 following the stocking of roach in 1993, a major die off of perch ≥ 2 years took place in both roach treatment lakes (LAT 2, 4). Population sizes of perch ≥ 2 years thereafter remained very low during 1995–97 in these lakes resulting in higher population numbers of perch ≥ 2 years in no roach lakes than in lakes with roach (repeated measure anovas 1993–97, Treatment, F1,2 = 18·1, P = 0·051, Time, F3,6 = 7·56, P = 0·018, treatment × time, F3,6 = 0·86, P = 0·51) (Fig. 1). After the drastic decrease of perch ≥ 2 years in 1994, the density of perch in LAT 2 has shown an increase starting from 1996 (regression, F1,5 = 8·96, P = 0·03) (Fig. 1a). In contrast, the density of perch ≥ 2 years remained low in LAT 4 up to 2003 when a drastic increase in the density of perch took place following a successful recruitment of perch in 2002 (Fig. 1b).

In one of the no roach lakes during 1993–96, LAT 3, the perch population underwent drastic shifts in population structure from a situation with a dominance of intermediately sized individuals to a dominance of small recruiting individuals and large giant individuals and thereafter back to a dominance of intermediately sized individuals (Persson et al. 2003, 2004) (Fig. 1b). No significant effect of the roach stocking in 1997 on this perch population has been observed and the density of perch ≥ 2 years in this lake has never decreased to the low levels observed in LAT 2 and 4 subjected to the roach stocking in 1993. Finally, in the other no roach lake during 1993–96, LAT 1, the density of perch ≥ 2 years has changed relatively little during 1994–2002, and no effect of the roach stocking in 1997 has been observed (regression, F1,5 = 0·75, P = 0·43) (Fig. 1a).

Overall, strong effects of roach introductions on perch ≥ 2 years were thus observed in LAT 2 and 4 but not for LAT 1 and 3. This stronger effect of roach introduction on perch ≥ 2 years in Lakes AT 2 and 4 compared with Lakes AT 1 and 3 was related to that the perch at the time of introduction was smaller in the two former lakes (median size 1-year-old perch excluded: one-tailed t-test, t2 = 3·4. P = 0·04, proportion of individuals > 180 mm 1-year-old perch excluded: one-tailed t-test, t2 = 3·2, P = 0·04).

roach introduction and performance

Following the introduction of roach into LAT 2 and 4, gill net caches were low up to 1998 (5 years after stocking) when numbers increased substantially in both lakes (Fig. 2a). The increase in numbers was related to the successful recruitment of roach born in 1997 evidenced by high numbers of roach in the 40–100 mm size category over the next 3(4) year period (years 5–7(8)) in the two lakes (Fig. 2b). In contrast to the stockings of roach in LAT 2 and 4, no successful recruitment of roach has been observed in LAT 3 even after 8 years and in LAT 1 first after 6 years, causing a clear effect of stocking year on roach invasion (repeated measure anova, test including years 1–7 after stocking, treatment factor time of stocking, F1,2 = 49·4, P = 0·02) (Fig. 2b). (Note that the few catches of roach in the size range 40–100 mm caught during years 1–3 in LAT 1 and 3 were results of capture of stocked small roach, see Table 1). The difference between LAT 2 and 4 vs. LAT 1 and 3 in roach recruitment pattern caused a lower overall capture of roach in the latter lakes (repeated measure anova, test including years 1–6 after stocking, treatment factor time of stocking × year, F1,5 = 19·4, P = 0·0001) (Fig. 2a). Most importantly, there has been no evidence for an increase in roach numbers in LAT 3 and the numbers caught have been very low from year 6 onward (Fig. 2a). The most recent data from 2005 using two sampling methods (trap and fyke net catches in spring, gill net catches in August) confirms both a total lack of recruitment in LAT 3 (only large roach originating from the stocking captured) as well as that a successful recruitment of roach has occurred in LAT 1 (Fig. 3a,b) as was indicated by previous data (Fig. 2b).

Figure 2.

(a) Catch per unit effort (CPUE, numbers) of roach in multimesh gill nets in the four lakes in years following roach stockings. For LAT 2 and 4, year 1 is 1994 and for LAT 1 and 3, year 1 is 1997. (b) Numbers of roach in the size interval 40–100 mm captured in traps and fyke nets during the May–June samplings in the four lakes in years following roach stockings. For LAT 2 and 4, year 1 is 1994 and for LAT 1 and 3, year 1 is 1997.

Figure 3.

(a) Total trap and fyke net catches of roach of different size classes in the four lakes in the spring of 2005. (b) Catch per unit effort (CPUE, numbers) of roach of different size classes in multimesh gill nets in the four lakes in August 2005.

With the exception of one year in LAT 1 and 3, respectively, YOY roach were reproduced in all lakes in all years and no effect of stocking year on the number of initial YOY roach numbers was present (repeated measure anovas years 1–6, Year of stocking: F1,2 = 2·58, P = 0·25, year of stocking × year: F3,6 = 0·21, P = 0·69) (Fig. 4). This lack of an effect of stocking year may, however, be due to high within-treatment compared with between-treatment variation leading to a low power due to low replication (cf. LAT 1 vs. LAT 3, Fig. 4). An inspection of Fig. 4 shows that for LAT 3 initial (after hatching) numbers of YOY roach have always been less than 1000 individuals ha−1 in all years and no survival after 5 weeks has been observed. This pattern corresponds to the complete absence of recruitment in this lake. For LAT 2 and 4, the recruitment of roach 5 years after stocking (Fig. 2b) corresponds to high initial numbers and high survivals of YOY roach in year 4 (Fig. 4). At the same time, high initial densities have not necessarily meant a high recruitment in these two lakes (e.g. LAT 2 years 2 after stocking, LAT 4 years 6 after stocking) (Figs 2 and 4). For LAT 1, high initial densities of YOY roach were present in years 1, 5 and 6 after stocking (Fig. 4). Substantial survival and recruitment the following years were, however, only present in years 5 and 6 after stocking (Figs 2b and 4).

Figure 4.

Start (at hatching) and final (5 weeks after hatching) densities (mean ± 1 SE) of YOY roach in the different lakes in different years in LAT 1–4 based on Bongo trawling.

Successful recruitment of roach measured both in terms of total number and recruitment (numbers of roach 40–100 mm) has thus been observed in LAT 2 and 4–5 years after stocking and in LAT 1 7 years after stocking. These successful recruitment events have all been associated with high reproductive output and a high early survival of roach (Fig. 4). A low reproductive output can explain the lack of successful recruitment in LAT 3. At the same time, not all high reproductive outputs in LAT 1, 2 and 4 have resulted in successful recruitments. Overall, patterns in roach larvae densities thus suggest that high reproductive output was a necessary but not sufficient condition for successful recruitment. In the following, two potential mechanisms to explain the presence of successful recruitment of roach in LAT 1, 2 and 4 and its absence in LAT 3 will be investigated: resource limitation, and predation pressure from perch.

resource limitation

Because strong recruitment pulses of YOY perch have been shown to be able to heavily depress the shared resource zooplankton (Persson et al. 2003, 2004), a possible explanation for the lack of successful invasion of roach in LAT 3 may be related to more severe resource limitation. This hypothesis was tested by calculating consumption rates based on observed resource levels for two size classes of YOY roach (30 and 50 mm) of which the smaller size corresponds to the length of roach in the middle of July and the larger size corresponds to the maximum length observed at the end of the growth season.

Estimates of resource limitation showed that both size classes of roach generally had consumption rates very close to their maximum consumption rate. For 30 mm roach, consumption rates in LAT 2 and 4 were in 11 cases out of 14 above 80% of the maximum consumption rates (10 of 14 cases for 50 mm roach) (Fig. 5). For LAT 1 and 3, consumption rates for both size classes of roach were in 13 of 14 cases above 80% of the maximum consumption rates. No difference in the degree of resource limitation was found between stocking years (repeated measure anovas years 1–6, 30 mm roach, year of stocking: F1,2 = 0·55, P = 0·54, year of stocking × year: F3,6 = 0·92, P = 0·51, 50 mm roach, anovas years 1–6, year of stocking: F1,2 = 0·56, P = 0·53, year of stocking × year: F3,6 = 0·71, P = 0·63). Furthermore, the degree of resource limitation was overall lowest in LAT 3 where no successful recruitment has been observed with an estimated consumption rate always exceeding 90% of the maximum (Fig. 5).

Figure 5.

Consumption rates of two size classes of roach as a proportion of the maximum consumption rates in the four lakes in different years starting with the year of stocking (1993 for LAT 2 and 4, 1996 for LAT 1 and 3). Dashed horizontal lines denote consumption rate necessary to meet maintenance requirements.

Estimates of the degree of resource limitation thus suggest (1) an overall low degree of resource limitation, and (2) a higher degree of resource limitation in the year with successful recruitment of roach into LAT 2 and 4. These findings are in agreement with comparisons of observed mass increases with maximum mass increases in YOY roach over the whole first growth season (Persson et al. 2000). Both the analysis based on resource levels presented here and the analysis based on growth rates thus strongly suggest that resource limitation cannot explain the differences found in recruitment between lakes.

predation by perch

Following the first roach stocking in 1993, mature perch suffered from a very heavy mortality in LAT 2 and 4 as described above leading to negligible reproductive outputs in 1995 and 1996 (Fig. 6a). This resulted in an absolute absence of OYO perch in 1996 and 1997 (years 3 and 4 after roach stocking) in these lakes (Fig. 6b). In 1997, reproduction in LAT 2 and 4 was again present as a result of high growth (and per capita fecundity) of the remaining individuals leading to an increase in the number of OYO perch the following year. From 1998 and onwards, perch reproduction was present every year in LAT 2 and 4 resulting in substantial numbers of OYO perch (Fig. 6b).

Figure 6.

(a) Mean densities over the first 5 weeks after hatching (± 1 SE) of YOY perch in the different lakes for 6 years following the roach introduction (LAT 2 and 4, 1994–1999, LAT 1 and 3, 1997–2002). (b) Trap and fyke net catches of OYO perch in the different lakes for 6 years following the roach introduction (LAT 2 and 4, 1994–1999, LAT 1 and 3, 1997–2002). (c) Perch population attack rate (size-specific attack rate × population numbers summed over 5-mm size cohorts) of perch in the different lakes in 6 years following the roach introduction (LAT 2 and 4, 1994–1999, LAT 1 and 3, 1997–2002).

In contrast to the roach stockings in 1993, mature perch did not suffer significantly from the roach stockings in 1996 (Fig. 1). In LAT 3, continued reproduction every year resulted in that substantial numbers of OYO perch were present in all years although cannibalism heavily reduced the number of YOY perch over the growth season (Persson et al. 2004). In LAT 1, YOY perch were also present in every year, but a drastic decrease in the number of OYO perch took place in 2000 (year 4 after roach stocking) followed by two additional years with low OYO perch numbers (Fig. 6b).

Estimates of perch population attack rates summed over all size classes show that the perch population attack rate on roach varied considerably over time in three of the four study lakes. For the two lakes where roach were stocked in 1993, perch population attack rate on roach decreased to very low values in years 3 and 4 after the roach stocking to thereafter increase again (Fig. 6c). The decrease in perch population attack rate was a result of the negligible reproduction in 1995 and 1996 following the high mortality of adult perch (Figs 1 and 6a). The negligible reproductive output, in turn, resulted in the absence of OYO perch the following 2 years, causing the very low estimated predation intensities in years 3 and 4 after roach stocking (Fig. 6b,c). In these two lakes, the observed successful recruitment and increase in roach numbers in 1997 (Fig. 2) is thus clearly related to a time window with low predation intensity in years 3 and 4 after the roach stocking. YOY roach survival during their first month was consequently very high in 1996 and 1997 in these lakes (> 50% years 3 and 4 after stocking) (Fig. 4).

In LAT 3, perch population attack rate was high over the whole study period either as a result of high predation intensity from the stunted mature perch individuals or from OYO perch (Fig. 6c). Correspondingly, roach population densities have been very low in this lake over the whole study period and YOY roach have been consistently reduced to zero in all years during their first month (Fig. 6). In LAT 1, roach reproductive output was high in both 2001 and 2002 (years 5 and 6 after roach stocking) and substantial numbers of YOY roach remained in the lakes 1 month after hatching (Fig. 4). Estimates of predation intensity show that the predation intensity decreased years 4–6 after roach stocking (2000–02) (see also number of OYO perch trapped, Fig. 6b). Patterns in survival rate of YOY roach and estimates of perch predation intensity thus suggest that a time window with low predation intensity was present in 2000–02, and that successful recruitment of roach was observed in LAT 1 in the last two years of the study (Fig. 2b).


short-term recruitment success of roach

The results show that successful recruitment of roach has been observed in three of the four experimental lakes. Compiling the data on (1) initial density of roach larvae, and (2) predation pressure for all lakes shows that the presence of successful roach recruitment depended on both high initial YOY roach larvae density and low perch predation pressure (Table 2). In support of the fact that both conditions are necessary for successful recruitment, independent occurrences of either high initial densities (LAT 2 1995, LAT 4 1999) or low predation pressures (LAT 1 2000, LAT 2 1996, LAT 4 1996) were observed in three of the lakes. Still, the presence of only one of the two conditions has never resulted in successful recruitment (Table 2). Interestingly, roach population fecundity in one of the lakes that was stocked with roach on the later occasion (LAT 3) never reached high levels. The reason for this pattern may be that this lake had a substantial number of very large piscivorous perch in the year (1996) when the roach stocking was carried out (Persson et al. 2004) and these perch likely imposed a strong predation pressure on the smaller size category of the stocked roach.

Table 2.  Invasion success of roach. Occurrence (indicated by solid grey bars) of high initial densities of roach larvae (> 8000 individuals per ha), low predation pressure (perch population attack rate < 2 × 105 L day−1) and high roach recruitment (Fig. 3) the following year in the four lakes as a function of years since stocking
LakeYear since stockingHigh initial density of roach larvaeLow predationHigh roach recruitment next year
11 (1997)   
6 (2002)   
21 (1994)   
6 (1999)   
31 (1997)   
6 (2002)   
41 (1994)   
6 (1999)   

Different patterns in perch response to roach introductions were observed between the two introduction experiments in 1993/94 and 1996/97 as mature perch were heavily affected in LAT 2 and 4, but not in LAT 1 and 3. In LAT 2 and 4, the stocking of roach resulted in a reduction in the availability of two shared resources, zooplankton and macroinvertebrates, and in body condition and growth of perch compared with the control lakes at this time (LAT 1 and 3) causing a massive death in both mature as well as 1-year-old perch in LAT 2 and 4 (Byström et al. 1998; Persson et al. 1999). The reduced performance and mortality of perch following the roach introduction is consistent with the hypothesis that roach induces a competitive juvenile bottleneck in the recruitment of perch to larger piscivorous stages (Werner & Gilliam 1984; Persson & Greenberg 1990; Olson et al. 1995).

The absence of an effect of roach on perch ≥ 2 years in LAT 1 and 3 can be related to differences in competitive effect of roach on perch in the different lakes as perch were larger in LAT 1 and 3 than in LAT 2 and 4 at the time of the roach stocking. Consequently, perch were expected to share fewer resources with roach due to size-dependent resource use (Persson 1988). In LAT 3 specifically, at the time of the roach stocking, the perch population was dominated by large perch. These large perch were not expected to be affected by roach competition as the perch gained most of their energy from cannibalizing on YOY perch (Persson et al. 2003, 2004).

The data for all four lakes consistently showed that a low predation pressure was necessary for a successful recruitment of roach. Three major mechanisms can be advanced to have caused the reduced perch predation pressure. First, because perch population numbers can fluctuate due to intrinsic mechanisms (Persson et al. 2003) it can be hypothesized that successful roach recruitment may take place in time windows with low population numbers. The data do, however, not lend any support for this hypothesis as the predation pressure of the widely fluctuating perch population in LAT 3 remained high over the entire study period as evidenced by population attack rate estimates. Second, roach may, as discussed in the previous section, impose an interspecific competitive effect on perch leading to reduced perch numbers and thereby to reduced perch predation pressure on roach. This mechanism evidently caused the reduced predation pressure from perch in LAT 2 and 4 (Fig. 6). Third, the presence of pike in two of the experimental lakes (LAT 1 and 2) resulting in overall lower densities of perch (Fig. 1) (Persson et al. 1996) may facilitate recruitment in roach. The consistently lower perch population attack rates over study years in LAT 1 compared with LAT 3 (Fig. 6) lend support to this hypothesis.

life-history omnivory and long-term persistence of roach in perchroach systems

The perch–roach system represents a life-history omnivory (IGP) system where large perch prey on small roach, whereas small perch compete with roach for zooplankton. Furthermore, our previous experiments (Persson 1987; Persson & Greenberg 1990; Byström et al. 1998) have demonstrated a competitive superiority of roach. In IGP systems such a competitive superiority of prey has been shown to be crucial for the existence of the intermediate consumer (roach) under any environmental condition (Diehl & Feissel 2000; Mylius et al. 2001). Modelling analyses using perch and roach parameters also suggest that the perch–roach system represents a tightly connected size-structured IGP system (van der Wolfshaar et al. 2006). Previous multigenerational experimental studies of IGP systems are restricted to laboratory microcosms (Morin 1999; Diehl & Feissel 2000, 2001; Price & Morin 2004; see also Krivan & Diehl 2005 for an overview). These microcosm experiments manipulated productivity levels of the systems allowing a test of IGP theory with respect to expected changes in trophic structure with productivity. In the whole lake experiments, it was for obvious logistic reasons not possible to do this. At the same time, it was possible to measure a number of critical factors for IGP theory like roach reproductive output, roach resource limitation and perch predation impact (Figs 4–6). This circumstance makes it relevant to address the question about the expected coexistence of perch–roach in the experimental lakes in relation to IGP theory particularly in the lakes inhabiting only perch at the start of the experiment (LAT 3, 4).

In a nutshell, analyses of IGP models reveal four ranges of system productivity with different predictions about community composition. At low productivities the top predator cannot persist, either because productivity is too low for its existence even in the absence of the intermediate consumer or because it is outcompeted by the intermediate consumer despite the fact that system productivity is sufficient for its persistence. Coexistence of top predator and intermediate consumer and alternative stable states (top predator-intermediate consumer–resource state or top predator–resource only state) may occur at intermediate productivities. At high productivity, however, the intermediate consumer is excluded through predation (Holt & Polis 1997; Diehl & Feissel 2000; Mylius et al. 2001). Food-dependent development in predator and intermediate consumer dramatically decrease the scope for coexistence between size-structured top predator and intermediate consumer in IGP models (Van der Wolfshaar et al. 2006), although bistability between two alternative stable states, either an intermediate consumer–resource or a top-predator–resource equilibrium, is still predicted to occur at intermediate productivities, With respect to these four ranges of productivity we can first conclude that the productivity of the experimental lakes was clearly not too low to allow the existence of perch even when roach are at low numbers, as perch were present in all four lakes before the roach introductions. Second, the inability of roach to invade LAT 3 and the recovery of perch in LAT 4 argue against a scenario that the productivity of the lakes is in the range where roach excludes perch by competition. Third, it is not likely that the lake productivities were in the range with alternative stable states of either a roach–resource or a perch–resource community in light of the development of perch in LAT 4. In this lake the perch population was reduced to very low levels (c.15 individuals ≥ 2 years in the whole lake), and the roach population reached very high numbers (Fig. 2).Nine years after the roach introduction, the perch population was nevertheless back to very high numbers pointing to an ability of perch to increase from very low numbers in the presence of high numbers of roach. Taken together, the data rather support the fourth scenario where perch excludes roach through predation, and the dominance of roach over a number of years in LAT 4 may therefore have only been transient and long-term extinction of roach is expected.

pike as a factor mediating coexistence between roach and perch

The density of perch in the lakes with pike present was lower than in the other two lakes (Persson et al. 1996). Owing to the large scale of the whole lake experiments, the degree of replication was by necessity limited, but it may nevertheless be hypothesized that pike presence may be important for coexistence between perch and roach. Pike prey on both perch and roach, but show a positive selection for perch over roach (Persson et al. 1996; Nilsson & Brönmark 2000). It can therefore be argued that pike presence may mediate coexistence between roach and perch through selective predation on perch. Comparative data lend support for this hypothesis. In a survey of 32 Finnish lakes, Sumari (1971), found that roach were only found with perch when together with other piscivores, in most cases pike (14 of 32 lakes). Notably, the situation that roach and perch do coexist in many lakes is in agreement with this expectation as perch and roach are generally found together with other fish species and particularly with pike (Persson et al. 1991; Jeppesen et al. 1997; Haertel et al. 2002).

In conclusion, the whole lake experimental results lend support to the fact that coexistence between size-structured predator and size-structured intermediate consumer is limited due to the predation effect from the predator, and that other interacting species may be essential for long-term sustenance of the intermediate consumer. Increased lake heterogeneity with lake size may, in addition to other top predator presence, also be hypothesized to promote coexistence due to reduced resource overlap between predators and intermediate consumers resulting from spatial segregation.


We thank Magnus Huss, Emma Lindgren, Johan Lövgren, Krister Matsson, Karin Nilsson, Anders Persson, Lena Staffans, Stefan Sjögren, Rickard Svanbäck, Eva Wahlström, Roger Wallin and Erika Westman for field and laboratory assistance during different phases of the study. We also thank the Åman Fishery Co-operative for access to the study lakes. Valuable comments that lead to a major restructuring of the paper were given by anonymous reviewers. The study was funded by the Swedish Research Council and the Swedish Research Council for Environment, Agricultural Sciences and Spatial Planning to L. Persson. A.M. De Roos was supported by the Netherlands Organization for Scientific Research.