Introduction

Shallow lakes in The Netherlands are inhabited by large freshwater bivalves of the family of Unionidae, such as Anodonta anatina L. (pond mussel). We were interested to see if A. anatina may contribute—like the zebra mussel Dreissena polymorpha Pallas (Reeders & Bij de Vaate, 1990)—to the improvement of water quality by reducing algal biomass and increasing water transparency. Enhanced grazing on phytoplankton may help shift shallow lake ecosystems with high and intermediate nutrient loading from a turbid, phytoplankton dominated, to a clear, water state dominated by macrophytes (Scheffer, 1998). One reason to predict a substantial impact of Unionid bivalve grazing on ecosystem processes is that, in marine and estuarine systems, similarly sized bivalves have been shown to have large ecosystem impacts on the benthic-pelagic coupling of energy cycles (Prins et al., 1998). A second reason to focus on Unionid species for the control of the effects of eutrophication is the potential risk of zebra mussel spreading in ecosystems. Although zebra mussels—with their high filtration rates on phytoplankton (Fahnenstiel et al., 1995)—have been promoted as a tool in biomanipulation of lakes in the Netherlands (Reeders & Bij de Vaate, 1990), they have also been shown to promote blooms of (toxic) cyanobacteria in North America (Vanderploeg et al., 2001). Moreover, we believe that for Dutch shallow lakes, with soft substrates, Unionids are better adapted to these habitats than zebra mussels that require hard substrate for settlement.

We studied the feeding of A. anatina on cultures of the green alga S. obliquus Turp. Kütz., toxic and non-toxic strains of the filamentous cyanobacterium P. agardhii Gomont and colonial M. aeruginosa Kütz., each at five different concentrations to determine the functional response. The functional response is the relation between feeding rate and food density and as such may provide indications of underlying mechanisms, such as an increase in handling time related to the size of food particles. When knowing the functional response of A. anatina for a certain type of food, we know more about which mechanism may be limiting to the food uptake and hence we can predict how the grazer may respond to this food type in the field. In bivalves particles are first filtered from the water column. In the mantle cavity (where gills and palps are situated), ingestible particles are directed straight to the mouth, whereas unwanted particles are sorted on the gills and palps, embedded in mucus and then expelled as pseudofaeces (PF) (Dionisio Pires, 2005).

Because the functional response of Anodonta has never been analysed before, and different types of functional responses have been described in bivalve literature, we compared three basic types of functional responses (Holling, 1959; Hassell, 1978). The simplest type, type I (rectilinear) functional response, shows a linear increase of the ingestion rate (IR) with the food concentration, up to a concentration where the maximum amount of food that can be ingested by an organism is reached. This concentration is called the incipient limiting level (ILL: Rigler, 1961). Above ILL, the IR remains constant and maximal irrespective of food concentration. This implies that above ILL, further increases in cell concentration result in higher PF production rates (Sprung & Rose, 1988). The clearance rate (CR), a measure for the efficiency with which food is filtered from the water column, decreases accordingly (Fig. 1). Often, such a discontinuous transition at ILL is not observed, unless the food is filtered from the environment with negligible handling of the food. For many organisms, handling time of food is a limiting factor for the amount of food that can be ingested. In a type II (hyperbolic, curvilinear) functional response, this is visible in CR, which decreases progressively towards zero with increasing food concentrations. Consequently, IR decelerates at higher food concentrations and goes asymptotically towards the maximum ingestion rate. With a type III functional response, IR accelerates more than linearly at low food densities, which for example may be the case with reward-dependent feeding behaviour. The initial CR will be 0, and maximal at intermediate food densities (Fig. 1). Because IR will still saturate (and thus decelerate) at higher food concentrations, as in a type II functional response, this will result in a sigmoidal (S-shaped) functional response (Fig. 1). With a type III functional response, CR will be maximal at the half-saturation food concentration K (Fig. 1).

Fig. 1
figure 1

Graphic representation of the three basic types of functional responses for ingestion rates (IR) and clearance rates (CR), respectively, as a function of food concentration (F) (solid lines). The dotted vertical lines indicate for Type I functional response the incipient limiting level (ILL), whereas they indicate the half-saturation concentration (K) for Type II and Type III functional responses. For comparability, all functional responses are scaled to a maximum IR or CR of 1, and to and ILL or K of 1, on axes with arbitrary units. Therefore, for type I, IR is maximum at ILL, and (for types II and III) IR is at half its maximum value (0.5) at K. Note that—for comparability with experimental data—graphs are plotted on logarithmic scales, which makes it difficult to immediately recognize the sigmoid shape of the Type III functional response

Full size image

To investigate the functional response of A. anatina, we chose the different food items based on differences in food quality (green algae vs. cyanobacteria), shape (filamentous, colonial or single cells), and toxicity (toxic or non-toxic strains). Our null hypothesis (1) is that A. anatina has a negligible handling time of the food, meaning that it has a type I functional response. Therefore, (2) production of PF starts only above ILL and increases linearly with food concentration. We expect that A. anatina is able to discriminate between green algae and (often less nutritious) cyanobacteria so that: (3) CR on S. obliquus is higher than on all the cyanobacteria. Furthermore because retention efficiency is species specific and depends on particle size (Ward & Shumway, 2004); we expect that (4) small species like M. aeruginosa are more easily retained than large filaments of P. agardhii, which may congest the siphon. Finally (5) with respect to toxicity (i.e. microcystin-LR), CR and IR on microcystin producing cyanobacterial strains will be lower than on non-toxic strains. Our study adds new data on the functional response of the rarely studied freshwater bivalve A. anatina feeding on different food sources. These data are relevant for studies of behaviour and populations, as well as for the possible application of these bivalves in biomanipulation of turbid shallow lakes.

Materials and methods

Grazing experiments

Anodonta anatina mussels were collected in spring 2005 in the Babbelaar (Lake Lauwersmeer, The Netherlands). Mussels were kept in aquaria filled with filtered lake water (0.45 μm, Lake Maarsseveen) on a layer (10 cm) of river sand (Aqua-colisa, Gemert-Bakel). Mussels were kept at 17–20°C, under a light: dark regime of 16:8 h and fed every day with S. obliquus CCAP276/3A. Water in the aquaria was replaced once a week. Five phytoplankton strains were used in the experiments: four cyanobacteria (P. agardhii strains CYA116 and CYA126 and M. aeruginosa strains V131 and V40) from exponentially growing batch cultures, and one green alga (S. obliquus CCAP276/3a), which was cultured in a chemostat. For size characteristics, origin and toxicity of the strains we refer to Table 1. Before the experiments, the M. aeruginosa strains were filtered over a 70 μm filter, to remove extremely large colonies.

Table 1 Characteristics of the phytoplankton strains used in the experiments
Full size table

Grazing experiments were performed in the laboratory, with A. anatina feeding on the five phytoplankton strains as single food sources. For each strain, the mussels were exposed to a range of five food concentrations (0.5; 1; 2; 5; 10 mg C L−1), with 5 replicates per strain per concentration. Food concentrations were based on regressions between ash-free dry weight of the strains (4 h at 550°C) and optical density at 750 nm (Helios-δ, Unicam, UK). Per grazing vessel, one mussel, with an average shell length between 8 and10 cm and a shell width between 5 and 6 cm, was cleaned with a brush under running tap water to remove phytoplankton adhered to the shell and incubated in 2 l of the appropriate food concentration (suspended in 0.45 μm filtered lake water). The grazing vessels were 2 l batch systems and contained no sediment, as a pilot study showed that sediment absence did not affect the grazing behaviour of A. anatina.

Acclimation with the appropriate food concentrations started 24 h prior to the actual grazing experiment (19–20°C; 10–15 μmol photons m−2 s−1). Two hours before the grazing experiment food was added again to ensure that the mussels were well fed before the actual start of the experiment (Dionisio Pires et al., 2005). The 2 l were enough to prevent complete food depletion during the second acclimation and the grazing experiment period. Before the actual grazing period, the mussels were rinsed carefully under running tap water. We included duplicate vessels for each of the food concentrations without mussels as controls to correct for changes in the phytoplankton biomass other than those related to grazing. During the experiment the vessels were constantly mixed by aeration to keep food in suspension, without disturbing the mussels. Samples (15 ml) were taken from all vessels (both mussels and controls) at the start (T0 min), before the introduction of the mussels, and at the end (T60 min) of the experiment. After the grazing period the mussels were rinsed under running tap water and transferred to a vessel with 0.45 μm filtered lake water and samples were taken from these vessels after 30 min (T90 min) to assess PF production. Samples were stored in the dark, until being measured at the end of the experiment. Measurements (chlorophyll concentrations (μg l−1)) were done using a PhytoPAM (Walz, Germany). Carbon concentrations of the phytoplankton were measured on a UniQuant Universal Carbon and Nitrogen Analyzer (Calanus, Finland). After the grazing period, the mussels were stored at −20°C. At a later stage, the shells were removed, the soft tissue was freeze-dried, and weighed.

Determination of clearance rates, PF production rates and ingestion rates

Mass-specific clearance rates (CR, ml mg DW−1 h−1) were determined from chlorophyll fluorescence measurements and calculated as follows (Coughlan, 1969):

$$ {\text{CR}} = \frac{V} {{nt}}*\left\{ {\ln \frac{{A_0 }} {{A_t }} - \ln \frac{{A_0'{} }} {{A_t'{} }}} \right\}, $$

in which V is the volume in the grazing vessel (2,000 ml), n the dry weight of a single A. anatina mussel (mg), t the duration of the experiment (in h), A 0 the algae concentration (μg chlorophyll l−1) in the vessels with mussels at t = 0, A t the algal concentration in the vessels with mussels at time t, A 0′ and A t ′ the concentration of algae in the control vessels at t = 0 at time t, respectively. Note that each replicate contained one animal. The average algal carbon concentrations ( \(\overline A \) mg C l−1) were determined by multiplying the geometric mean of initial and final algal concentrations with the slope (s) of previously made regressions for each phytoplankton species between carbon and chlorophyll concentrations:

$$ \bar A = s \cdot \sqrt {A_t \cdot A_0 } $$

Slopes of the regressions between chlorophyll and carbon were all forced through 0, and all had r 2 values higher than 0.99. The average algal carbon concentrations and clearance rates were used to calculate gross mass-specific ingestion rates (Lürling & Verschoor, 2003) (IRg, mg C mg DW−1 h−1):

$$ {\text{IR}}_{\text{g}} = \bar A \cdot {\text{CR}} $$

To calculate net ingestion rates, we first calculated mass-specific PF production rates (PPR, mg C mg DW−1 h−1) from the algal concentrations in the PF production treatments (A P):

$$ {\text{PPR}} = \frac{{s \cdot V \cdot A_{\text{P}} }} {{\Delta t \cdot n}} $$

so that net mass-specific ingestion rates (IR, mg C mg DW−1 h−1) could be calculated by subtracting PPR from gross ingestion rates:

$$ {\text{IR}} = {\text{IR}}_{\text{g}} - {\text{PPR}} $$

For ease of comparison among different treatments, individuals, and food concentrations, we also calculated PF production as a fraction of gross ingestion (F P):

$$ F_{\text{P}} = \frac{{{\text{PPR}}}} {{{\text{IR}}_{\text{g}} }} $$

Because for one strain (S. obliquus CCAP276/3A) PF data were available for only three concentrations (0.5, 1, and 5 mg C l−1), we pooled the data for only these concentrations within each phytoplankton strain. Due to non-normality and heteroscedacity of these data, even after arcsin transformation, we applied nonparametric tests for comparisons among these strains. Overall differences of these fractions (F P) between the different strains were compared using the Kruskall–Wallis test, followed by pairwise (two-tailed) Mann–Whitney U-tests for comparisons between strains.

For S. obliquus CCAP276/3A, the original PF data were considered to be too limited in terms of number of data points and range of food concentrations to make reliable model fits of the functional response models. Therefore, the average of these F P values was used to estimate the net ingestion rates from gross ingestion rates.

Fitting and intercomparison of functional response models

Functional response models were fitted by iterative nonlinear regression of functional response models on mass-specific ingestion and clearance rates. Because variance increased with food concentration, we minimized the squared residuals between log10-transformed observations and model predictions. The functional response model that best fitted to a particular data set (i.e.: leaving the smallest squared residuals) was used for comparisons between phytoplankton strains.

The type I (rectilinear) functional response fitting functions used were:

$$ {\text{if }}\bar A < {\text{ ILL then IR}} = {\text{IR}}_{\max } \frac{{\overline A }} {{{\text{ILL}}}},{\text{ and if }}\bar A > {\text{ ILL then IR}} = {\text{IR}}_{\max } , $$

with ILL being the incipient limiting level: the food concentration where the ingestion rate becomes constant maximum, and IRmax being the maximum ingestion rate. Clearance rates were fitted similarly, and for reasons of comparability, maximum clearance rates (CRmax) were calculated as derived parameters:

$$ {\text{if }}\bar A < {\text{ ILL then CR}} = \frac{{{\text{IR}}_{{\text{max}}} }} {{{\text{ILL}}}} = {\text{CR}}_{\max } ,{\text{ and if }}\bar A > {\text{ ILL then CR}} = \frac{{{\text{IR}}_{\max } }} {{\bar A}} $$

Type II (curvilinear) fitting functions used were

$$ {\text{IR}} = {\text{IR}}_{\max } \frac{{\bar A}} {{\bar A + K}},\quad {\text{CR}} = \frac{{{\text{IR}}_{\max } }} {{\bar A + K}}\;\;{\text{so}}\;\;{\text{CR}}_{\max } = \frac{{{\text{IR}}_{\max } }} {K}{\text{, }} $$

with K being the half-saturation food concentration at which half of IRmax is reached, which is a measure for the initial slope and curviness of the functional response. With low K, the functional response starts steeply and saturates rapidly with increasing concentration, whereas at high K, the functional response rises and saturates slowly, and at very high K the functional response is practically linear.

Type III (sigmoid) fitting functions used were

$$ {\text{IR}} = {\text{IR}}_{\max } \frac{{\bar A^2 }} {{\bar A^2 + K^2 }},\;\;\;{\text{CR}} = {\text{IR}}_{\max } \frac{{\bar A}} {{\bar A^2 + K^2 }},\;\;\text{so} \;\;{\text{CR}}_{\max } = {\text{IR}}_{\max } \frac{{{\text{IR}}_{\max } }} {{2K}}, $$

with K being the half-saturation food concentration, as well as the inflection point of the sigmoid curve, and the food concentration at which CRmax is reached.

For the comparisons of CRmax and IRmax between phytoplankton strains, we performed pairwise comparisons between each possible strain pair. Our null hypothesis (H0) for comparisons was that the separate functional response models were not significantly better predictors then when the model was fitted to the pooled observations for both strains. This hypothesis was compared against the alternative hypothesis (H1) that the separate models were better predictors, using the significance of the maximum likelihood ratio statistic G (also known as G 2, Bishop et al., 1975),

$$ G = 2N \times \ln \left( {\frac{{{\text{MS}}_{{\text{among}}} }} {{{\text{MS}}_{{\text{within}}} }}} \right) $$

with N being the total number of observations (i.e. samples), MSamong the total variance of the residuals between the log10-transformed predicted and observed values of the alternative model (H1), and MSwithin the variance of the residuals of the null model (H0). G has a χ2 distribution with degrees of freedom numbers corresponding to the difference between the numbers of parameters of the two models being compared.

Results

Clearance rates

CR decreased with increasing food concentrations, except for toxic M. aeruginosa (V40) where the CR was essentially constant (Fig. 2). The residuals, indicating the goodness of fit of the different functional response models (Table 2, lower indicates better fit), showed that clearance rates on S. obliquus (CCAP276/3A), both P. agardhii strains (non-toxic CYA116 and toxic CYA126) and non-toxic M. aeruginosa (V131) yielded a type II response. For toxic M. aeruginosa both type I and II fitted equally well (similar residuals) (Table 2). CRmax decreased in the order S. obliquus > toxic P. agardhii > non-toxic P. agardhii > toxic M. aeruginosa > non-toxic M. aeruginosa (Fig. 2, Table 3). K is extremely large (4.7 × 1012) for toxic M. aeruginosa (Table 3), meaning no decrease in CR at higher concentrations, and indicating that a linear (type I) functional response applies here. However, for reasons of comparability between strains, we used the type II functional response model with this extreme value for K, which yields a practically linear function. K is high for non-toxic M. aeruginosa and S. obliquus, and low for non-toxic and toxic P. agardhii (Fig. 2, Table 3). The functional responses based on clearance rates were all significantly different from each other, except for between toxic and non-toxic P. agardhii (Table 4).

Fig. 2
figure 2

Mass-specific clearance rates (CR) of Anodonta anatina grazing on different concentrations of five different phytoplankton strains. Food concentrations are geometric means of initial and final concentrations. Dots represent measured clearance rates, solid lines represent Type II functional response model regressions, dotted lines represent Type I functional response model regressions

Full size image
Table 2 Residual sum of squares for the fits of the three different functional response models to data of clearance rates (CR) and ingestion rates (IR) of Anodonta anatina on five different phytoplankton strains
Full size table
Table 3 Parameter estimates of nonlinear regression of type II (curvilinear) and type I (rectilinear) functional response models of clearance and ingestion rates of Anodonta anatina on five different phytoplankton strains
Full size table
Table 4 Pairwise comparisons of the best-fitting functional response models of clearance and ingestion rates of A. anatina on five different phytoplankton strains
Full size table

Ingestion rates

On S. obliquus and both P. agardhii strains (non-toxic CYA116 and toxic CYA126), A. anatina had a type II response (Fig. 3 and Table 2), but for both M. aeruginosa strains (non-toxic V131 and toxic V40) a type I response was a better fit (smaller residuals for I than II; Table 2). For the type II functional response, K values as well as IRmax decreased in the order toxic M. aeruginosa > non-toxic M. aeruginosa > S. obliquus > non-toxic P. agardhii > toxic P. agardhii (Table 3), meaning that handling times were shortest for both M. aeruginosa strains (non-toxic V131 and toxic V40). For both M. aeruginosa strains, K and IRmax were smaller with type I than with type II functional responses, yet they were still larger than for non-toxic and toxic P. agardhii. Strains that differed significantly in their functional response based on CR also differed significantly in their functional responses based on IR (Table 4), and if a type I functional response would be fitted on both M. aeruginosa strain data, these responses would still be significantly different (Table 4).

Fig. 3
figure 3

Mass-specific ingestion rates (IR) of Anodonta anatina grazing on different concentrations of five different phytoplankton strains. Food concentrations are geometric means of initial and final concentrations. Dots represent measured ingestion rates, white triangles are estimated ingestion rates (see Materials and methods), solid lines represent Type II functional response model regressions, dotted lines represent Type I functional response model regressions

Full size image

Pseudofaeces production

For all phytoplankton strains, PF production increased more than linearly with increasing food concentrations, and we did not observe absence of PF production at low food levels (i.e. below the incipient limiting level). The Kruskall–Wallis test revealed significant differences in PPR by A. anatina, when grazing on the five different strains (p < 0.0001, H = 30.41, d.f. = 4, N = 64). Pairwise comparisons (Mann–Whitney U-test) showed that A. anatina had significantly lower PPR for toxic M. aeruginosa, compared to the other cyanobacterial strains (p < 0.05), but PPR was not significantly different from S. obliquus (p > 0.05, Fig. 4). Significantly higher PPR were found for non-toxic M. aeruginosa than for S. obliquus (p < 0.05).

Fig. 4
figure 4

Fraction of PF production relative to gross ingestion (F P) of Anodonta anatina grazing on five different phytoplankton strains. Bars represent averages, error bars represent standard deviations

Full size image

Discussion

The type I (rectilinear) functional response is often thought to be applicable for ‘true’ filter feeders such as bivalves, which is illustrated by the frequent use of the term ‘incipient limiting level’ in the literature. Unfortunately, such terminology is often not supported by a full functional response analysis. In our experiments, we found that A. anatina had a type II (curvilinear, hyperbolic) functional response on 4 out of 5 investigated phytoplankton strains, indicating that the handling time of the food plays an important role in the ingestion process. Therefore, we reject hypothesis 1 (clearance and ingestion rates are according to a type I functional response) for all strains except for the toxic M. aeruginosa (V40).

The striking exception to the observed functional responses is the CR on the toxic M. aeruginosa, both on the basis of the functional response type (type I) and our parameter estimates. These suggest that toxic M. aeruginosa should be very small but in fact this strain consists of single cells and intermediately sized colonies (Table 1). Although these colonies are easily retained on the gills, maybe even better than the smaller sized single cells, ingestion of such particles would still involve some handling time. Visual observations of the colonies of toxic M. aeruginosa showed that these are not solid and packed in a thick mucus layer, as is often observed in large M. aeruginosa colonies but loose aggregates that disintegrate easily. It is not difficult to imagine that such colonies could disintegrate due to shear forces in the feeding process. Thus, the actual size of the particles once ingested could be much smaller, as small as single cells, leading to very short handling times. The combination of easy capture on the gills (of intermediately sized aggregates) and easy ingestion (of small cells) could thus explain why we found a type I functional response of A. anatina for this strain.

We did not observe absence of PF production at low food levels (i.e. below the incipient limiting level), so that hypothesis 2 is rejected as well. Since PF production increased more than linearly with food concentration, this again indicates that a type II functional response is more appropriate for the feeding behaviour of A. anatina. The differences in fractions of PF production relative to gross ingestion (Fig. 4) suggest that A. anatina is able to discriminate between food sources by varying the degree of excretion (Sprung & Rose, 1988). The high fractions of PF production for non-toxic M. aeruginosa (V131) (30% of gross ingestion) and both P. agardhii strains (non-toxic CYA116 and toxic CYA126) (15–20% of gross ingestion) indicate that these strains were the least preferred. Both PF and feeding data show that S. obliquus (CCAP276/3A) is the preferred food type, which is in concordance with our hypothesis 3.

Hypothesis 4, coccal/colonial M. aeruginosa is preferred over filamentous P. agardhii, is less easy to accept or reject. Maximum clearance rates were higher on both non-toxic and toxic filamentous P. agardhii than on non-toxic M. aeruginosa, and equal to toxic M. aeruginosa, showing that P. agardhii is filtered at least equally efficient from the water column at low food densities. On the other hand, maximum ingestion rates were higher for both non-toxic and toxic M. aeruginosa, which indicates that M. aeruginosa is preferred over Planktothrix at high food densities. Probably the higher maximum clearance rates on both P. agardhii strains are artefacts caused by the larger size of these algae, which makes it more likely that they are filtered by chance. The high PF production on non-toxic and toxic P. agardhii species suggests that they are not very suitable food sources. At higher concentrations, long handling times of the filaments rapidly becomes a serious limiting factor in the ingestion process, which is visible in the curviness of the functional response (Figs. 2, 3, low K values in Table 3). With M. aeruginosa, however, the functional responses did not reach satiation (Figs. 2, 3).

Thus, at very high food concentrations, the feeding of A. anatina was not limited anymore by the amount of food in the environment, but solely by the handling time of the food and the estimated maximum ingestion rates were the inverse of the handling time of the food. For the filamentous P. agardhii we expected long handling times (e.g. due to clogging of the siphon), and indeed this species yielded the lowest maximum ingestion rates of all (Table 3). In our study, biovolume (Table 1) was inversely related to the maximum ingestion rate (Table 3), which shows that food particle size can be used as a predictor of feeding rates.

Contrary to our hypothesis 5, toxicity did seem not seem to affect the grazing behaviour of A. anatina. For M. aeruginosa, the toxic strain is preferred over the non-toxic strain. Moreover, the higher maximum ingestion rate of the toxic M. aeruginosa, as compared to S. obliquus, suggests that toxin production did not interfere with filtration and ingestion. Previous findings describe that mussels are relatively insensitive to microcystin producing cyanobacteria (e.g. Dionisio Pires et al., 2004). We are unable to distinguish the effects of size (toxic P. agardhii twice as long as non-toxic P. agardhii (Table 1)) and toxicity of the toxic P. agardhii on the lower ingestion rates by A. anatina on this strain as compared to the non-toxic P. agardhii strain. However, because we did not find a significant effect of microcystin on the feeding behaviour of A. anatina in the Microcystis treatments, it seems more plausible that size rather than toxin caused the lower ingestion rate of A. anatina on the toxic Planktothrix strain.

For the discussion on the applicability of A. anatina as a biofilter, we compare the filtration rates of A. anatina (mean dry weight: 3.5 g) with that of Corbicula leana in the field (Hwang et al., 2004) and D. polymorpha in the laboratory (Dionisio Pires, 2005). Hwang et al. (2004) showed in the hypertrophic Lake Ilgam, (3 × 105–8 × 105 cells ml−1 of M. aeruginosa and 0.6–2.8 mg C l−1 of P. agardhii) that C. leana filtered 1.6–7.8 l mussel−1 day−1. Furthermore, field experiments with D. polymorpha by Reeders & Bij de Vaate (1990) in a Dutch lake revealed a filtration rate of 1.2 l mussel−1 day−1 (with a shell length of 18 mm). Comparing grazing capacities per unit biomass when offering the same phytoplankton species (concentration 2 mg C l−1), we found that A. anatina cleared 0.38; 0.22; 0.34; 0.22 and 0.42 ml mg−1 h−1, and D. polymorpha (mean individual dry weight of 9 mg; Dionisio Pires, 2005) 4.0; 4.6; 11.3; 3.4 and 3.1 ml mg−1 h−1 of S. obliquus (CCAP276/3a), non-toxic- and toxic P. agardhii (CYA116 and CYA126) and non-toxic- and toxic M. aeruginosa (<60 μm) (V131 and V40), respectively. This means that for the phytoplankton strains used in this study, the grazing capacity, on a weight-specific basis, of D. polymorpha is ∼7–33 times higher than that of A. anatina. Despite the low clearance per unit of biomass, the grazing capacities per individual of A. anatina per day (CRmax of 42.6; 33.5; 37.1; 21.3 and 33.1 l−1 day−1 of S. obliquus, non-toxic P. agardhii; toxic P. agardhii, non-toxic M. aeruginosa and toxic M. aeruginosa, respectively) were comparably high with those of marine bivalves (M. edulis 35.3 l mussel−1 day−1 and Cerastoderma edule (Linnaeus, 1758) 31.2 l mussel−1 day−1; Ward & Shumway, 2004). On a weight-specific basis, the grazing capacity of Anodonta is low compared to mussel species as D. polymorpha. However, per individual, Anodonta’s grazing capacity is equal or higher than other species.

The current abundance of Unionid bivalves in the field (0–5 mussels or 17,500 mg DW m−2, as found in a field study in four shallow lakes in The Netherlands during 2005, unpublished data) however, is too low for A. anatina to have a strong effect on algal biomass. In contrast, in high density areas of Lake IJsselmeer and Lake Markermeer, D. polymorpha has been described to remove ±70% of the algal population (Lammens, 1999). When we take into account the number of 1,000 individuals per m−2 of Dreissena observed in Lake IJsselmeer (Bij de Vaate, 1991), an average individual DW for D. polymorpha and A. anatina of 9 and 3,600 mg respectively, and the grazing rates for Anodonta found in this study and for Dreissena found in Dionisio Pires (2005), we would need 83 Anodonta mussels per m−2 to clear the same volume of water from e.g. the toxic P. agardhii (CYA126) (at a food density of 2 mg C l−1) as would 1,000 Dreissena mussels. However, densities of Unionid mussels in the field appears to be low (unpublished data). On the other hand, the observed high densities of D. polymorpha (individuals per m−2) 500–1,000 and 400–1,000 individuals per m−2 in Lake IJsselmeer and Lake Markermeer, respectively (Bij de Vaate, 1991); 30,000 individuals per m−2 in Lake Zürich; 20,000 individuals per m−2 in Lake Garda and 21,000 individuals per m−2 in Lake Constance (Burla & Lubini-Ferlin, 1976; Franchini, 1978 and Suter, 1982 in Bij de Vaate, 1991) make it more likely that these smaller mussels have a larger effect on the primary productivity and a more realistic achievement for biomanipulation. If A. anatina is to be used as a biofilter in shallow Dutch lakes, its densities must be promoted. Current densities may be low because after years of eutrophication the stability of the sediments has changed, providing unsuitable habitat for these mussels.

Summary and conclusion

From this study we conclude that A. anatina mussels can filter and ingest colonial as well as larger filamentous cyanobacteria. This observation indicates that A. anatina may be useful as a biofilter even when cyanobacteria blooms are toxic, provided that densities of these mussels in the field are increased through suitable restoration measures (to be developed).