Study site

Field experiments were conducted in two ponds; pond 1 (P1) 0.6 ha and one pond 2 (P2) 0.7 ha in size, at a fish farm in Lohmar, Germany (50°49'33.00''N 7°12'59.42''E; for more detailed description of the ponds see [16]). The ponds are connected via overflows, assuring that they did not differ in abiotic parameters, such as temperature, oxygen content and conductivity (t-test, all P >0.05; see [17]). The present study is part of a larger investigation on the development of cannibalism. In the context of this larger study, in spring, the ponds were stocked with adult perch and bream (not further considered in this study) in April 2006 (for further details see [17,18]). Before stocking, the ponds were fishless. Adult perch stocked originated from the same population to minimize genetic differences which might, for example, influence growth rates [19]. After spawning, adult fish were removed from the ponds to guarantee undisturbed development of the offspring. A few weeks after hatching some YOY perch first became early piscivores (preying on bream larvae), thereby benefiting from accelerated growth [18]. Shortly after becoming piscivorous, these perch were large enough to cannibalise their smaller conspecifics, which led to establishment of two size-cohorts, a large-bodied cannibalistic and a small-bodied plankti-/macroinvertivorous cohort [18]. At this time the present study started, which means that these two size-cohorts were already present at the beginning of this investigation. Perch density within the first 60 days after perch hatching (before this experiment) did not differ between ponds (t-test P >0.05). For further explanation on how densities were calculated see S1 File.

Sampling

Starting in mid-June, zooplankton and macroinvertebrates were sampled weekly and biweekly, respectively. After size measurement and counting of organisms, biomass was determined from published length–mass equations and expressed as mg wet weight L^-1 for zooplankton samples, g wet weight m^-2 for sediment macroinvertebrates and catch per unit of effort (CPUE) [g section^-1] for macroinvertebrates in the vegetation. For further details on sampling of zooplankton and macroinvertebrates refer to S2 File.

Perch were sampled monthly using electro-fishing and gillnetting with multi-mesh-sized gillnets (for further details see [16]). Immediately after capture fish were killed with an overdose of metomidate hydrochlorid (Aquacalm ®) and deep-frozen for further analysis. The experiments complied with the current laws of Germany and were approved by the ethic commission of the University of Cologne. All sampling procedures were reviewed and specifically approved as part of obtaining the field permit. The owner of the experimental ponds gave his permission to conduct our studies there. The field studies did not involve endangered or protected species. In the laboratory fish analysed were randomly chosen from all sizes of caught perch, kept separated and registered individually. The length (total length, TL) (to nearest 0.5 mm) of each perch was recorded and each perch was photographed for morphometric analysis using a digital camera [20]. Subsequently, stomach content was analysed. Perch with empty stomachs (4–19% of all perch analysed) were excluded from further analyses, leaving 38–70 individuals per sampling date. The food spectrum of each perch was expressed as the weight percentage composition of food items identified to genus level [21]. For perch of the last sampling date in October the sex and the stage of maturation were documented. As perch had been registered individually, it was possible to assign each photo and its morphometric information to the corresponding stomach content analysis and where applicable stage of maturation of individual perch.

Data analysis and statistics

For further analysis of stomach contents, prey items were assigned to three groups; zooplankton, macroinvertebrates and fish, and data were expressed as wet weight percentage composition of these three prey categories in perch stomachs. Differences in diet composition of perch were determined by the Bray–Curtis similarity index [22]; i.e., by hierarchical clustering, using the obtained percentage weight data [23]. Data were clustered through group average linking, after which significant differences between samples were calculated by the non-parametric test of similarity (one-way ANOSIM). To illustrate the clustering, an MDS (non-metric multidimensional scaling) plot was drawn for a similarity level of 30% and a stress of 0.01. All cluster analyses were done with PRIMER v6 (Plymouth Routines in Multivariate Ecological Research, Roborough, Plymouth, UK). The diet spectrum of fish assigned to one cluster consisted of at least 50% of the corresponding resource; e.g., fish which had fed on 51% macroinvertebrates were clustered into the macroinvertivorous cohort.

The mean TL ± standard deviation of fish in each cluster was computed. Difference in TL was tested using Student’s t-tests for June and July and one-way ANOVA and Bonferroni post hoc tests for pairwise comparison for the other sampling dates. Growth was calculated as the difference of mean length in each cluster per day [mm day^-1]. All statistical tests referring to sizes of perch were performed with SPSS 23 (IBM, Armonk, New York, USA).

For morphometric analysis photos of perch and tpsDigit and tpsUtility software by Rohlf (available at: http://life.bio.sunysb.edu/morph/) were used to digitize fourteen landmarks on the perch’s left side as well as a scale (for details see [20]). Perch were grouped according to their assignment to different clusters (c.f. cluster analysis above), after which the Integrated Morphometrics Package (IMP) developed by Sheets (available at: http://www3.canisius.edu/~sheets/morphsoft.html) was used for further morphometric analysis. The differences of shapes between groups (clusters) were analysed using a canonical variates analysis (CVA) and, based on the Wilk’s lambda value at a p < 0.05 level of significance, the CV axes were tested for significance. The shape change was graphically illustrated by vectors on landmarks. As fish differed in size, potential morphological differences were examined for allometric effects. If allometric effects could be detected, shapes were standardized by regressing each group on size separately, thereby removing the variance attributable to size and leaving the shape variation that was independent of size differences. For a more detailed description of the morphometric analysis, the standardisation process and the software used for each calculation; see [20].

Modelling approach

To further understand how divergent resource use allows a shift from the small-bodied to the large-bodied fraction within a bimodal size-distribution of YOY fish, and to what extent this pathway is open to the initially small individuals, a stage-structured matrix model for YOY fish was used to compute YOY dynamics. The, which is described in more detail in S3 File, model is intended only to demonstrate the mechanisms by which the shifts can occur, and not to attempt to rigorously predict the empirical data. To keep the model simple, we started with an initially small–bodied cohort. We modelled three resources, zooplankton, macroinvertebrates and prey-fish, and enabled diet shifts of YOY from one resource to another. The cohort is divided into 300 stages (weight classes) to obtain a high resolution as growth rate is sensitive to weight differences in YOY perch. Thus, sufficient resolution to describe the continuous variation of weights that are expected in a cohort, was guaranteed. Each stage was associated with a length and weight of the YOY fish in that stage class. These stages can be considered ‘micro-stages’, as they don’t correspond to distinct physiological stages, but only to incremental differences in length and weight. The probability of fish advancing from stage class to stage class in each time step is less than 1, and depends on the amount of available prey. For parameters used in the model see S3 File. The model was calculated with Matlab (MathWorks, Natick, Massachusetts, USA).

The cohort of YOY perch is divided into three consumer groups: planktivores (B1), macroinvertivores (B2), and piscivores (B3), with: where B1_i(t), B2_i(t), and B3_i(t) are, respectively, the numbers of individuals of the three consumer groups in each of the 300 stages. The simulation keeps track of the YOY through their growing season. These YOY start as purely planktivores (B1) with an initial stage distribution (length and weight). At some point a fraction attains a stage with great enough length to move into the macroinvertivore group (B2), and eventually some fraction of those grow to a large enough length to advance to the piscivore group (B3).

These stage-structured cohort dynamics for the three consumer groups is entirely governed by a Markov chain matrix model: where A, which describes unidirectional growth in size, is: where a_i,i+1 represents the fraction advancing from a given stage class i to age class i+1, and a_i,i represents the amount remaining in stage i. The elements a_i,i and a_i,i+1 differ depending on which of the three consumer groups the YOY are in at a given time step. The amount initially in stage i that advances to the next stage, i+1, during a time step is where surv is a fixed constant survival rate for each time step and advance(prey) is the fraction of the survivors that advance to the next stage class, which is a function of prey density. The fraction of those initially in stage class i that remain there during the next time step is For simplicity, we assume that the type of prey being consumed does not affect survival, surv, but that it does affect advancement, through the functions; where Z, F, and M are the current available biomasses of zooplankton, prey fish, and macroinvertebrates, respectively. The fraction of advance to the next stage depends on the density of available prey. The q_is are constants that can be chosen to corresponding to given assumptions on how the prey biomasses of each type affect advances to the next stage.

In this case it was assumed that YOY perch feeding on fish reach the highest growth rates, followed by those feeding on macroinvertebrates, while planktivorous perch have the lowest growth rates (i.e., q₃ > q₅ > q₁). Additionally, the growth rates also depend on the concentrations prey, which are changing through time. The q_i’s are calibrated to give realistic rates of growth. Prey densities are described by the dynamic equations: where Weight_i is the weight of perch in stage class i, f_z, f_m, and f_f are feeding rates, and 1/h is the maximum digestion of biomass per unit time by the perch. r_z, r_m, and r_f are renewal rates of zooplankton, macroinvertebrates and prey-fish, and K_z, K_m, and K_f are the carrying capacities. Finally, perch are assumed to be able to switch prey types when they reach certain threshold stage classes, and switch then with certain probabilities each time step. At this time there are only two switches, from planktivory to macroinvertivory and from macroinvertivory to piscivory.

1. Stage_transfer_to_inverts = length at which planktivores can start to switch to macroinvertivory
2. Stage_transfer_to_fish = length at which macroinvertivores can start to switch to piscivory
3. Fraction_transfer_to_inverts = fraction of planktivores switching to macroinvertivory each time step
4. Fraction_transfer_to_fish = fraction of macroinvertivores switching to piscivory each time step

Diet shifts and morphometry

The performed shifts match widely known descriptions of perch’s ontogenetic diet shifts, where perch change from a planktivorous via a macroinvertivorous to a piscivorous diet [8]. However, diet shifts are usually performed over several years (perch regularly become piscivorous when 2 or 3 years old, [8,10]), while in this case the first group of perch escaping from the small cohort performed both diet shifts within one growing season. Therefore, at times the diet of medium-sized perch significantly differed from that of their smaller and larger conspecifics.

For the first group of escapees the change in consumed food resources was reflected in significant shape changes of juvenile perch. Such changes in morphology are not only documented in the ontogeny of perch [28], but also in many other organisms. They are regularly attributed to changes in the preferred habitat [29], resource use [30,31] or predation pressure [32]. Morphological change could in this study first be detected in as short a time as 4 weeks [33]. Generally, the shift to a higher trophic position (first to macroinvertivory, then to piscivory) was reflected in the development of a deeper body and a shorter head (pelvic fins and operculum moving anteriorily), which was suggested to be the species-specific ontogenetic growth trajectory in juvenile perch, given a sufficient energy intake [20].

Competition and growth

The scenario of usage of the macroinvertivorous niche by the different size-classes and morphological forms (first and second group of escapees) provides some evidence that intra-specific competition may be an important factor in the whole process of the continuous formation of bimodality. The population density in the experimental ponds used for this study was high (5 Ind m^-2, S1 File, also see [18]) compared to other systems (e.g., about 0.6 Ind m^-2, [34]; about 0.3 Ind m^-2, [11]), and the small size-cohort represented the major part (more than 95%) of the YOY perch population. Additionally, zooplankton was soon depleted after perch hatching [18] and, hence, scarce during the period of this study (S2 File). High population density and low zooplankton resources without any doubt induced strong intra-specific competition within the small size-cohort [6], which is clearly reflected in low growth rates (on average 0.6 mm day^-1). At high levels of competition and low resource availability, fish were shown to shift to alternative food resources not used by conspecific competitors [6,35,36]. The food resource chosen differed among individuals, suggesting that some individuals are being favoured in terms of growth due to the fact that they feed on a more energetic resource (e.g. macroinvertebrates or fish). Consequently, such a process of resource use diversity with higher growth rates, which may accelerate within short periods [7], would be rather independent of ontogenetic diet shifts. Exactly such a scenario became obvious in this study, as size-differences within the small size-cohort were rather minor, but still only a part of all individuals escaped from the small cohort by shifting to higher trophic levels. On the other hand, and in contrast to zooplankton, macroinvertebrates were not depleted but appeared in amounts similar to (or even higher than) other ponds inhabited by fish (e.g. all over biomass 0.5–8 g m^-2 at 50 kg ha^-1 fish, [37]; this study 5–11 g qm^-2 [note that in this study only sediment organisms!] [18]).

After the macroinvertivorous niche was occupied by escapees from the small size-cohort, remaining small conspecifics solely consumed zooplankton (reflected in minimal growth rates of 0.5 mm day^-1), giving some evidence that perch that did not perform the diet shift were the weaker competitors for macroinvertebrates and hence were forced to consume the less profitable and still depleted resource zooplankton. In fact laboratory experiments using three-spined stickleback (Gasterosteus aculeatus L.) fed with two sizes of Daphnia demonstrated that stronger competitors can force subordinates to opt for less profitable food items [38]. On the other hand, the small number of perch that had shifted to macroinvertivory (escapees) now benefited from accelerated growth (maximal 1.4 mm day^-1), which might be associated with a lower level of experienced intra-specific competition for macroinvertebrates (as not all perch had shifted to macroinvertivory). This is corroborated by a study on largemouth bass (Micropterus salmoides Lacépède), indicating a growth advantage of early diet shifters (towards a piscivorous diet) compared to bass becoming piscivorous at later stages [12]. Later, in October, medium-sized perch had left the macroinvertivorous niche and had become piscivorous, thus completing a shift from the small to the large fraction within a bimodal size-distribution. Thereafter, most perch of the small-sized individuals shifted from a planktivorous to macroinvertivorous diet; thus, on their part, reducing experienced intra-specific competition. Consequently, this diet shift of remaining small perch reflects the fact that once competition is no longer influencing the small size-cohort, the pathway for growing into the large size-cohort was potentially cleared.

The stage-structured model supported the mechanism described above. At the beginning of the calculation small perch had only minor size-differences. However at day 15 modelling results showed that some of the small perch start incorporating macroinvertebrates into their diet. These individuals soon benefited from accelerated growth and were finally able to shift to piscivory. The pathway of shifting to macroinvertivory and later to becoming piscivorous, remained valid for other individuals of the small size-cohort. Hence, the model shows that once perch start shifting from macroinvertivory to piscivory, more and more small individuals follow and start feeding on macroinvertebrates; hence, on their part, accelerating growth and catching up in size with the large piscivorous cohort.

As known for many fish species, perch display sexual dimorphism in size, growth and maturation [39–41]. Female perch, for instance, were shown to invest in active feeding [42] to ensure somatic growth and later gonad development. Male perch need less energy to produce sperm and therefore are not as dependent on high energy intake and growth as females [43]. The pathway, as described above allowing fish via accelerated growth to switch from a small size-cohort to a large one could therefore be favoured by female perch. Our results showed that the ratio of males and females caught in October (females: 53.6%) corresponds well to the natural sex ratio of other perch lakes ([42] and references therein). The large piscivorous size-cohort, however, was slightly dominated by males (57.7%). There was no sex-specific difference of perch TL in the large size-cohort; however, almost all male perch were found to be mature (93.3%), while this was only the case for one female (9.1%). These results suggest that the switch of perch from the small to the large size-cohort accompanied by diet shifts and morphological changes is performed by both sexes, although males may grow into the large cohort slightly more often. While both sexes can reach the same size, males directly benefit from the switch into the large piscivorous cohort, as they can mature and thus reproduce after only one growth season, which was also shown in other studies [24].

Development of bimodality

The change of size variation over time based on differing in individual growth rates within a population has been shown for several animals across different taxa [2]. For instance, in tadpoles (Rana sylvatica LeConte), individual differences in foraging ability led to increased size variation within populations under high competition [15]. The origin of bimodality has been attributed to either initial size differences or to one discrete period of accelerated growth in one part of the population [1,12,13,44]. Initial size-differences involve various factors, such as maternal effects [45] or the timing of spawning ([3] and references therein). A competitive environment may enhance the importance of size-independent factors, leading to increased size variation over time being more likely under high competition [13,15].

Most studies on bimodality have shown that once two size-cohorts developed, bimodality further amplifies with time. This amplification was studied for bimodal size-distributions related to factors such as spatial variation in resource availability or size-dependence and temporal variation in availability of resources [2]. If resource availability increases with size, and temporal variations in resource levels exist, an initial size distribution easily develops into bimodality [46]. Size-dependent resource as the origin of bimodal size-distributions has been verified by theoretical approaches and also an amplification of bimodality through time has been demonstrated [1]. The results of former studies indicating the amplification of bimodality revealed that existing size-cohorts, once they have been formed, are rather discrete and individuals of a population do not grow from one cohort into the other. This is also corroborated by a long-term study on Arctic charr (Salvelinus alpinus L.) in a Norwegian lake demonstrating that the size-cohorts within a bimodal size-distribution, which had developed after a mass removal of stunted charr, could be tracked down for several years [47].

Generally, the mechanism described in our study, where differences in individual growth rates over time in a competitive environment led to the formation of distinct size-cohorts, resemble former studies examining the development of bimodality [18,24]. In addition, our study showed that the formation of a bimodal size-distribution is a continuous process persisting for at least a whole growth season. The fact that bimodality is not necessarily attributable to one defined incidence (such as different spawning periods leading to an initial size difference or one discrete accelerated growth period of one part of the population) reveals that alternative and more continuous pathways exist for the formation of bimodality. Thereby, the ratio of differently sized individuals within the population continually changes in relatively short time periods. Cohorts differ not only in size but also in morphological, physiological or behavioural traits [16,20,24,48], which transfers into differing positions within and impact on the whole system. Therefore, it is highly important to consider continuously changing size-distributions for a better understanding of ecological processes and links within an ecosystem.