J. Phycol.

46, 33–40 (2010)

 2009 Phycological Society of America
DOI: 10.1111/j.1529-8817.2009.00775.x

RELATIONSHIP BETWEEN FISH AND THE NUMBER OF HORNS IN

CERATIUM HIRUNDINELLA (DINOPHYCEAE): A FOOD-WEB-MEDIATED
EFFECT ON ALGAL MORPHOLOGY? 1

Andrea Bertolo2
Groupe de recherche sur les écosystèmes aquatiques, Département de chimie-biologie, Université du Québec
à Trois-Rivières C.P.500, Trois-Rivières, Québec G9A 5H7, Canada

Ge´rard Lacroix, Françoise Lescher-Moutoue´

Ecole Normale Supérieure, UMR-CNRS 7618 (Bioemco), Biogéochimie et écologie des milieux continentaux,
46 rue d’Ulm, 75230 Paris Cedex 05, France

and Joakim Hjelm

National Board of Fisheries, Institute of Marine Research, Turistgatan 5, Box 4, SE-453 21, Lysekil, Sweden

In a freshwater mesocosm experiment, we
explored the potential for direct and indirect
effects of roach (Rutilus rutilus) and Eurasian perch
(Perca fluviatilis), two planktivorous fishes with dif-
ferent feeding behaviors, on the morphology of
Ceratium hirundinella (O. F. Müll.) Dujard., a large
dinoflagellate. Three morphs were detected: one
with two hypothecal horns, one with a third rudi-
mentary horn, and one with three well-developed
horns. We observed a strong negative relationship
between the presence of fish and the proportion of
three-horned cells. The two fishes had strikingly sim-
ilar effects on C. hirundinella morphology, despite
their different capabilities to retain particles of the
size of C. hirundinella. This finding suggests that the
morphological variation in C. hirundinella was not
related to selection by fish. Morphological variations
in C. hirundinella could not be explained by fish-
mediated variations in turbidity (i.e., light climate)
or by predation pressure by the fish. In contrast, the
proportion of three-horned cells was directly related
to the biomass of filter-feeding cladocerans. This
result was unexpected since cladocerans are not
considered to consume C. hirundinella and they did
not depress C. hirundinella numbers in our experi-
ment. Without excluding other possible mecha-
nisms, we suggest that the third horn might help
these dinoflagellates avoid physical contact with the
filtering apparatus of the cladocerans and the conse-
quent potential damage caused by these herbivores,
which were more abundant in the absence of plank-
tivorous fish.
Key index words: cladocerans; dinoflagellates; fish
predation; invertebrate predators; morphological
variation
1
Received 8 December 2008. Accepted 19 August 2009.
2
Author for correspondence: e-mail andrea.bertolo@uqtr.ca.
Abbreviations: dwt, dry weight; fwt, fresh weight; k,
light attenuation; RDA, redundancy analysis coef-
ficient; VIF, variance inflation factor
C. hirundinella is a large (150–250 lm total length)
freshwater mixotrophic dinoflagellate protected by a
rigid cellulose armor (lorica) (Huber-Pestalozzi
1950, Popovský and Pfiester 1990). Because of these
morphological characteristics, C. hirundinella is well
protected from ingestion by herbivorous cladocer-
ans (Sommer et al. 2003). When present in high
numbers, it may inhibit the feeding activity of large
cladocerans by interfering with their food gathering
and ingestion (Xie et al. 1998). In contrast, omnivo-
rous rotifers (i.e., Ascomorpha spp. or Asplanchna
spp.) and copepods (calanoids and cyclopoids) may
consume C. hirundinella (Santer 1996, Stelzer 1998,
Sommer et al. 2003).
C. hirundinella can be classified into three mor-
phs: one with two hypothecal horns, one with a
third rudimentary posterior horn, and one with
three well-developed horns (hereafter 2, 3), and 3+,
respectively; see fig. 2 in Hamlaoui et al. 1998).
Different authors (references in Hutchinson 1957)
reported that the number of horns and the body
size of C. hirundinella could vary seasonally. Such
cyclomorphosis has been mainly interpreted as an
adaptation to seasonal changes in water density and
viscosity because of temperature changes (e.g.,
Hutchinson 1957). However, Hamlaoui et al. (1998)
observed large variations in C. hirundinella morphol-
ogy in experiments with no change in water temper-
ature, suggesting that the appearance of the third
posterior horn could have other adaptive advanta-
ges. Recent studies (Kim et al. 2003, Lürling 2003)
suggested that the observed phenotypic variability in
different phytoplankton taxa could be related to the
presence of zooplankton grazers or planktivorous

33
34 ANDREA BERTOLO ET AL.

fish. We tested this hypothesis by analyzing the
results of a mesocosm study on the effects of plank-
tivorous fish on planktonic food webs (Bertolo et al.
1999a,b, 2000) in which C. hirundinella was observed
to be morphologically variable. We analyzed the
relationships between the abundance and morphol-
ogy of C. hirundinella and the factors listed below:
1. Turbidity. Experimental results indicate that vari-
ations in the light spectrum can modify the mor-
phology of C. hirundinella (Hutchinson 1957). We
might thus expect a link between turbidity and
the proportions of the different C. hirundinella
morphs.
2. Consumption by planktivorous fish. The morph
with three horns might be retained more effi-
ciently by fish gills than the morph with two
horns. Therefore, fishes characterized by con-
trasting capture efficiencies of small-bodied prey,
such as roach (R. rutilus) and Eurasian perch
(P. fluviatilis), might induce different morpholog-
ical patterns in C. hirundinella.
3. Predation by invertebrate predators. The pres-
ence of a third horn might reduce the probabil-
ity of capture or ingestion by predatory rotifers
and copepods. In this case, we would expect a
positive relationship between the biomass of
invertebrate predators and the proportion of the
3+ horned morph of C. hirundinella.
4. Mechanical interference with grazers. As stated
above, different studies showed that filter-feeder
zooplankton interact with C. hirundinella even
though it is not edible. A positive relationship
between grazers biomass and the proportion of
the 3+ horned morph would lend support to this
point.
MATERIALS AND METHODS
Experimental design. Experiments were conducted in Lake
Créteil (Lake Créteil, 484693799 N, 22694799 E; 15 km SE of
Paris, France) in 18 experimental enclosures of two different
sizes, shallow (1.5 · 1.5 · 2.5 m depth) and deep (1.5 ·
1.5 · 4.5 m depth). To minimize seston heterogeneity, enclo-
sures were filled randomly and on several occasions between
June 29 and July 3, 1995; the plankton communities in the
enclosures were allowed to develop prior to the application of
the treatments, which began on July 13 (day 0) and ended on
August 1, 1995. Thermal profiles were similar among the
enclosures during the experiment. Two fish species were used
in the experiment, roach (total length: 9.0 ± 0.4 cm, mean ±
SD) and Eurasian perch (6.0 ± 0.5 cm) (hereafter ‘‘perch’’).
Six different treatments were applied randomly to enclosures
with three replicates (Table 1): dC (deep – control), sC
(shallow – control), dR (deep – roach), sR (shallow – roach),
dcR (deep – caged roach), dP (deep – perch). The diversity of
treatments (see Bertolo et al. 1999a,b, 2000) induced highly
heterogeneous environmental conditions and the decoupling
of potential key factors (Table 2). Moreover, the comparison of
selected groups of treatments gave complementary information
on the four factors of interest (Table 3).
Turbidity: Previous analyses of treatments dC, sC, dR, and
sR showed strong additive effects of both roach predation and
shallowness on the standing crop of algae and turbidity
(Bertolo et al. 1999a). The presence of roach induced a 2-fold
increase of phytoplankton biomass through a top-down mech-
anism, while the reduction of enclosure depth induced a 7-fold
increase of phytoplankton biomass via a reduction of light
limitation (Table 2). The analysis of these four treatments gives
us the opportunity to decouple the effects of fish and turbidity
by comparing systems with the same type of food web (two or
three trophic levels) that differed in turbidity (i.e., dC vs. sC
and dR vs. sR).
Consumption by planktivorous fish: We compared the effects
of perch and roach (dP vs. dR) to understand the role of fish
consumption on C. hirundinella morphology. Perch is more
visually oriented and selects larger particles than roach (Hjelm
and Johansson 2003). Measurements of distances between gill
rakers and the size of prey ingested by perch and roach
indicated that 9 cm roach are more likely than 6 cm perch to
ingest particles 200 lm in length (J. Hjelm, unpublished data).
This finding suggests that the two fishes used in our experi-
ment differed in their ability to prey on C. hirundinella, and in
their potential to affect the morphology of this dinoflagellate.
Predation by invertebrate predators and mechanical interference
with grazers: Perch and roach had different effects on the
structure of the zooplankton communities (Table 2; Bertolo
et al. 1999a,b, 2000). This gave us the opportunity to decouple
the effects of invertebrate predators, herbivorous cladocerans,
and fish on C. hirundinella morphology.
In this study, we present only those results relevant to
understanding the patterns observed for C. hirundinella mor-
phology. For more details, see Bertolo et al. (1999a,b, 2000).
Sampling and analysis. We analyzed C. hirundinella mor-
phology on day 19 because previous enclosure experiments
conducted at the same site during summer (Hamlaoui et al.
1998) revealed a clear morphological response of C. hirundi-
nella a few weeks after fish introduction. Integrated samples
representing the whole water column were collected in

Table 1. Description of treatments used in the mesocosm experiment.

Treatment
code Full name Details

dC Deep – control 4 m deep enclosures without fish

sC Shallow – control 2 m deep enclosures without fish
dR Deep – roach 4 m deep enclosures with six roach (20 g fwt Æ m)2)
sR Shallow – roach 2 m deep enclosures with six roach (20 g fwt Æ m)2)
dcR Deep – caged roach 4 m deep enclosures with six caged roach (20 g fwt Æ m)2). Roach were
confined inside two net bags (200 lm mesh size), allowing fish excretion
but preventing predation on plankton outside the cages. Living
zooplankton netted in the pelagic zone of the lake were regularly added
to the bags for fish to feed on.
dP Deep – perch 4 m deep enclosures with 18 perch (20 g fwt Æ m)2)
fwt, fresh weight.
 INDIRECT EFFECTS ON CERATIUM MORPHOLOGY 35

Table 2. Treatment mean (± SD) and analysis of variance (ANOVA) results on zooplankton biomass and body size, chl a,
light attenuation coefficient (k), and Ceratium hirundinella abundance. Each treatment had three replicates. Treatments:
dC, deep – control; sC, shallow – control; dR, deep – roach; sR, shallow – roach; dcR, deep – caged roach; dP, deep –
perch. Asterisks indicate significant P-values after sequential Bonferroni correction. For each dependent variable, means
with the same letter are not significantly different (post hoc Tukey’s test).

Mean ± SD F5,12 P

dC sC dR sR dcR dP
)1
Biomass (lg dwt Æ L )
Cladocerans 80.7 ± 8.1b 192.0 ± 64.0a 24.7 ± 4.8c 41.1 ± 16.8c 86.2 ± 5.7b 2.0 ± 0.5d 120.9 <0.0001*
Cyclopoids 53.5 ± 1.4a 62.9 ± 29.3ab 83.4 ± 33.8a 56.1 ± 26.7ab 33.3 ± 4.3ab 19.0 ± 7.6b 4.8 0.0120*
Calanoids 112.3 ± 44.9a 145.5 ± 29.1a 164.8 ± 69.8a 123.0 ± 40.2a 130.6 ± 29.8a 1.9 ± 1.4b 49.8 <0.0001*
Rotifers 13.8 ± 3.6ab 18.8 ± 1.8ab 29.7 ± 5.9a 15.7 ± 3.5ab 10.3 ± 4.7b 40.8 ± 32.5a 4.3 0.017*
Asplanchna 0.2 ± 0.1bc 4.5 ± 1.5a 1.0 ± 0.7ab 3.1 ± 2.9ab 0 ± 0c 0 ± 0c 24.1 <0.0001*
Mean body size (lm)
Cladocerans 517 ± 28ab 555 ± 89a 417 ± 33bc 314 ± 26cd 460 ± 16ab 300 ± 17d 26.6 <0.0001*
Cyclopoids 546 ± 51ab 538 ± 48ab 457 ± 32b 603 ± 45a 526 ± 65ab 406 ± 62b 5.5 0.0072*
bc a bc abc ab c
Calanoids 681 ± 80 851 ± 113 606 ± 50 746 ± 49 708 ± 13 506 ± 66 8.8 0.0010*
)1  c b bc a c b
Chl a (lg Æ L ) 1.9 ± 0.6 14.1 ± 3.1 4.7 ± 1.4 30.9 ± 5.2 2.3 ± 0.4 7.5 ± 1.9 52.1 <0.0001*
k (m)1) )0.79 ± 0.01c )0.69 ± 0.01ab )0.80 ± 0.03c )0.58 ± 0.05a )0.75 ± 0.02bc )0.70 ± 0.07ab 14.9 0.0001*
C. hirundinella 155 ± 153bc 266 ± 151ab 94 ± 43bc 1168 ± 287a 174 ± 89bc 243 ± 144ab 5.6 0.0068*
(ind Æ L)1)

ANOVA performed on data that were log transformed.

ANOVA performed on data that were rank transformed.
dwt, dry weight.

Table 3. Summary of predictions on the variation in both the abundance and morphological variation of Ceratium hirundi-
nella. Bold indicates predictions supported by our data.

Predictionsa
Treatments
Factor considered Method C. hirundinella abundance C. hirundinella % of 3+ morph

Turbidity dC vs. sC ANOVA — dC „ sC

dR vs. sR — dR „ sR
Direct consumption by fish dP vs. dR ANOVA dP > dR dP > dR
dC vs. dcR vs. (dC, dcR) > dR > sR (dC, dcR) > dR > sR
dR vs. sR
Predation by invertebrates All Correlation and RDA 1 ⁄  cyclopoid biomass  cyclopoid biomass
1 ⁄  Asplanchna biomass  Asplanchna biomass
Mechanical interference All Correlation and RDA —  cladoceran biomass
with cladocerans
Treatments: dC, deep – control; sC, shallow – control; dR, deep – roach; sR, shallow – roach; dcR, deep – caged roach; dP,
deep – perch; RDA, redundancy analysis.
a
Dash indicates cases where no specific predictions were made.

each enclosure. Water samples were screened through GF ⁄ C
(Whatman International, Maidstone, UK) glass fiber filters
(nominal cut-off: 1.2 lm), and chl a content was measured
spectrophotometrically (Spectronic 501 spectrophotometer;
Milton Roy, Co., Rochester, NY, USA) after grinding the
filters in acetone (AFNOR 1990). The PAR flux (lmol Æ m)2
Æ s)1) for the estimation of the light attenuation coefficient (k,
m)1) was measured at the surface and at 1 m depth with a
GaAsp bleu model Quantum Sensor (University Paris XI,
France). Zooplankton and C. hirundinella were filtered
through a nylon screen with a 32.5 lm mesh size (Mougel
SAS, Thouaré-sur-Loire, France) anesthetized with carbonated
water, and then fixed in 4% formalin. Counts and measure-
ments of zooplankton and of C. hirundinella were made under
a stereomicroscope (Wild M8; Leica Microsystems, Wetzlar,
Germany) on four to five subsamples at different dilutions. If
few individuals were present, the whole sample was counted.
We did not use the classical Utermöhl method for counting
the three morphs of C. hirundinella because the number of
cells in most cases would have been insufficient to allow
precise analyses.
Zooplankton biomass was calculated by genus as in Bertolo
et al. (2000). Four cladoceran genera were present in the
enclosures: Ceriodaphnia, Bosmina, Diaphanosoma, and Daphnia.
Potential predators of C. hirundinella were the cyclopoids
Acanthocyclops and Thermocyclops, the calanoid Eudiaptomus,
and the rotifers Asplanchna. In the statistical analyses, we used
these genera separately or we grouped them into calanoids
(adults + copepodites), cyclopoids (adults + copepodites),
cladocerans, and rotifers. The morphology of C. hirundinella
was determined on at least 100 individuals per sample.
Statistical analysis. One-way analysis of variance (ANOVA)
followed by Tukey’s test was used for univariate comparisons.
Planned contrasts were used to test specific hypotheses. When
necessary, data were log(x + 1) or square-root transformed to
normalize the residuals. Percentage data were arcsine trans-
formed prior to analysis. If these procedures were not success-
ful, ANOVAs were conducted on rank-transformed data
(Quinn and Keough 2002).
We used redundancy analysis (RDA) (CANOCO program,
version 4.51; ter Braak and Smilauer 2002) to relate variations
in abundance of the three morphs to variations in zooplankton
36 ANDREA BERTOLO ET AL.

genera biomass and turbidity (k). Multivariate dependent data

were Hellinger-transformed before the RDA (Legendre and
Gallagher 2001). The variables that contributed most to
explaining the variations were selected using a forward
selection procedure with a cutoff point of a = 0.10. However,
in the final RDA model, we focused on those variables that were
significant at the more conservative a level of 0.05. The
statistical significance of both selected variables and ordination
axes (a = 0.05) were tested by a Monte Carlo permutation test
(n = 9,999). The collinearity between explanatory variables was
controlled by eliminating those variables with a variance
inflation factor (VIF) >10.
Fig. 1. Percent of 2 (with two hypothecal horns; open bars),
3) (with a third rudimentary hypothecal horn; hatched bars),
RESULTS and 3+ (with three well-developed hypothecal horns; filled bars)
Ceratium hirundinella morphs in the different treatments at day
The different treatments significantly affected the 19. s, shallow enclosures; d, deep enclosures; C, control (no fish);
abundance of C. hirundinella (Table 2). The highest R, roach; cR, roach confined into net bags; P, perch. Error bars =
abundances were found in treatment sR, and these ±SD, n = 3.
were significantly greater than abundances in dR,
dC, and dcR (Tukey’s test, a = 0.05). Treatments dP r = 0.50; P = 0.033). However, neither correlation
and sC showed intermediate values and were not was significant at the Bonferroni-corrected a level of
significantly different from any of the other treat- 0.003 (16 univariate pair-wise comparisons were
ments. On average, C. hirundinella abundance was made).
greater in shallow than in deep enclosures (ANOVA The redundancy analysis was significant
contrast: sC, sR vs. dC, dR; F1,16 = 18.8, P = 0.001); (P = 0.0005), and the first axis explained a large
this was also observed for total phytoplankton bio- percentage (89.9%) of the variation in the propor-
mass (i.e., chl a) (ANOVA contrast: F1,16 = 200.1, tions of different morphs (Fig. 2). The second axis
P < 0.0001). C. hirundinella abundance was 1.7 times explained only 0.8% of the variation. The first axis
greater in shallow (sC) than in deep (dC) enclo- clearly separated morph 2 (on the right) vs. morphs
sures with no fish, whereas it was 12.5 times greater 3) and 3+ (on the left) by separating free-ranging
in shallow enclosures with roach (sR) than in deep fish enclosures from the other ones.
ones (dR). This effect of enclosure size on C. hirun- The stepwise selection procedure in the RDA
dinella abundance began as early as 13 d (day 0) model retained the following variables (listed in the
after the enclosures were filled. At day 0, the abun-
dance was 3.2 times greater in shallow (1,966.3 ±
419.0 ind Æ L)1; mean ± SD) than in deep enclo-
sures (604.2 ± 229.0 ind Æ L)1; mean ± SD) (ANOVA
on a random subsample of enclosures taken before
fish were introduced; F1,15 = 31.2, P = 0.0025).
The different treatments had a very strong impact
on the morphology of C. hirundinella (Fig. 1). The
proportion of morph 3+ was significantly reduced in
the presence of free-swimming fish (F5,12 = 30.1,
P < 0.0001; sC, dcR, dC > dR, dP, sR; Tukey’s test,
a = 0.05). We did not find any significant difference
between (i) deep and shallow treatments, (ii) free-
swimming perch and free-swimming roach treat-
ments, (iii) control and caged roach treatments.
Pooling data for the 3+ and 3) morphs led to a simi-
lar result (F5,12 = 72.1, P < 0.0001; sC, dcR, dC > dR,
dP, sR). Both ANOVAs conducted on C. hirundinella
morphology were significant at the Bonferroni-cor-
rected a level of 0.025.
The total abundance of C. hirundinella was not
associated with any of the zooplankton taxa, whereas Fig. 2. Redundancy analysis (RDA) ordination diagram show-
ing the effects of zooplankton biomass and turbidity (k) on the
it was significantly correlated with the light attenua- numbers of 2, 3), and 3+ Ceratium hirundinella morphs on day 19.
tion coefficient k (Spearman’s r = 0.65; P = 0.0048). Dashed arrows represent the vectors of the selected explanatory
These results did not change qualitatively if only variables. Solid lines indicate the approximate location of depen-
large zooplankters (>500 lm) were used in the anal- dent variable centroids. The axis scaling is 1 for dependent and
explanatory variables and 0.42 for sites (i.e., enclosures). Bold
ysis (results not shown). indicates those explanatory variables significant at the 0.05 a-level.
Among the zooplankton taxa, only cyclopoids and s, shallow enclosures; d, deep enclosures; C, control (no fish); R,
calanoids were moderately correlated (Spearman’s roach; cR, roach confined into net bags; P, perch.
 INDIRECT EFFECTS ON CERATIUM MORPHOLOGY 37

same sequence as they were included in the model

with the fraction of the total variation they
explained): Ceriodaphnia (P = 0.0008; 43.7% of the
variation), Daphnia (P = 0.0003; 31.5% of the
variation), Bosmina (P = 0.074; 4.8% of the varia-
tion), k (P = 0.056; 5.0% of the variation),
Asplanchna (P = 0.058; 3.5% of the variation), and
Diaphanosoma (P = 0.086; 2.5% of the variation).
Ceriodaphnia and Daphnia, the only two variables sig-
nificant at the 0.05 a level, jointly accounted for
83% of the total variation explained by the RDA.
The forward selection procedure did not retain
copepod biomass (neither calanoids nor cyclopoids)
as explanatory variables in the model. The biomass
of the four cladoceran genera was mostly associated
with the negative part of the first axis, while
Asplanchna was mostly associated with the second
axis. The attenuation coefficient (k) was mostly asso-
ciated with the second axis and, to a lesser extent,
with the positive part of the first axis. The ordina-
tion clearly showed that morphs 3+ and 3) were
associated with an increase in the biomass of cladoc-
erans, while morph 2 was negatively associated with
this factor. In contrast, greater values of k seemed
only weakly related to an increase in morph 2 num-
bers. Given that the first axis separated treatments Fig. 3. Proportions of 3+ morphs (a) and abundance (ind Æ L)1)
with free-swimming fish (dR, sR and dP; positive (b) of Ceratium hirundinella versus the biomass of cladocerans
part of the axis) from treatment in which fish were on day 19. Spearman’s rank correlation coefficients and associ-
absent or caged (dC, sC and dcR; negative part of ated P-values are shown for each panel. Symbols as in Figure 2.
the axis), we ran two separate RDAs on these groups
of treatments. Although this approach reduced both
DISCUSSION
sample size and the gradient in zooplankton bio-
mass (and thus the power of the test), it allowed us Without excluding that other factors, such as
to analyze the effects of zooplankton and k indepen- water temperature, could also affect the morpholog-
dently of fish occurrence. The first RDA (dR, sR, ical variability of C. hirundinella, our results clearly
and dP) was significant (P = 0.011), and the two confirm that fish can have a strong impact on the
first axes together explained 61.8% of the variation proportions of the different morphs in the absence
in the proportions of different morphs (respectively of any thermal variability. The effects of fish
47.0% and 14.8%). The stepwise selection proce- observed in our experiment are strikingly similar to
dure retained only Daphnia (P = 0.0333; 39.7% of those reported by Hamlaoui et al. (1998), even
the variation) and Ceriodaphnia (P = 0.0396; 22.1% though the experimental designs were different
of the variation). Both genera were associated with (e.g., size and species of fish used). In contrast to
greater proportions of morphs 3) and 3+ (positive the experimental design used by Hamlaoui et al.
portion of the first axis). The second RDA (dC, sC, (1998), our experiment allowed us to better analyze
and dcR) was only marginally significant (P = 0.064) the relationship between the morphological varia-
and had only one canonical axis, which explained tion and the different potential factors, such as the
31.7% of the variation in the proportions of the effects of fish and invertebrate predation, herbivo-
different morphs. The stepwise selection procedure rous zooplankton, and turbidity, without confound-
retained only Daphnia (P = 0.064; 31.7% of the ing these factors (Fig. 4). Our results are admittedly
variation), which was associated with greater correlative; nevertheless, we can offer different lines
proportions of morph 3+ (positive portion of the of evidence to support the notion that the observed
first axis). pattern of C. hirundinella morphology is caused by
Univariate comparisons showed that the pro- food-web-mediated effects and not by direct fish pre-
portion of morph 3+ was positively associated with dation.
total cladoceran biomass (Spearman’s r = 0.82; Effects on C. hirundinella abundance. C. hirundinella
P < 0.0001; Fig. 3a). The proportion of morph 3+ abundance was greater in shallow enclosures than
showed a shift at 70 lg dry weight (dwt) Æ L)1 of in deep enclosures, as was also observed for the
cladocerans. No significant correlation was found total phytoplankton biomass. Light was probably
between the proportion of the 3+ morph and the an important factor controlling phytoplankton
other potential consumers of C. hirundinella or k. biomass in our enclosures (Bertolo et al. 1999a):
38 ANDREA BERTOLO ET AL.
Fig. 4. Schematic representation of the relationships among
different factors considered in this study and their effects on the
relative abundance of three-horned Ceratium hirundinella cells, as
suggested by our mesocosm experiment. Continuous and dotted
arrows represent positive and negative effects, respectively. Arrow
thickness is proportional to the intensity of the effect.
phytoplankton in shallower enclosures were not
light limited and thus increased in biomass. The
positive, significant correlation between k and
C. hirundinella abundance could thus be viewed as
a bottom-up effect, which also affected the other
phytoplankton taxa.
In contrast, we did not find any clear pattern
between the variation of C. hirundinella abundance
and the abundance of fish. These results are at odds
with those of Hamlaoui et al. (1998), who deter-
mined that the abundance of C. hirundinella
increased along a gradient of 0+ roach density
(0, 10, 20, or 40 individuals per enclosure) in 4.5 m
deep enclosures identical to ours. On the basis of
our results and those by Hamlaoui et al. (1998), we
have no evidence of a direct control of C. hirundinella
abundance by roach or perch. Moreover, the abun-
dance of C. hirundinella cells was not correlated to
predator or herbivorous zooplankton abundance,
suggesting that this species was not controlled by
zooplankton-induced mortality in this experiment.
Effects on C. hirundinella morphological variation.
Turbidity: The RDA showed that the proportions of
the different C. hirundinella morphs were only
weakly related to variations in turbidity. The contri-
bution of this variable to the explained variation
was not significant and was small compared to the
contribution of cladocerans. The variations in fish
biomass and turbidity were confounded only within
a given enclosure depth (i.e., deep or shallow) in
our study, but these two factors were only weakly
correlated when all the treatments were considered
together. Thus, there is little support for the
hypothesis that fish affected the morphological
variation in C. hirundinella via a top-down effect on
turbidity in our experiment.
Consumption by planktivorous fish: Perch and roach
differ in their ability to capture and retain small par-
ticles. While it is possible that both the 3.0 cm roach
used in the experiment by Hamlaoui et al. (1998)
and the 9.0 cm roach used in our experiment con-
sumed C. hirundinella, it seems very unlikely that
6.0 cm perch consumed it, based on their poor
capability to retain small particles (J. Helm, unpub-
lished data, Nunn et al. 2007). Despite these differ-
ences, perch and roach induced strikingly similar
morphological responses in C. hirundinella in both
experiments. Hence, the hypothesis of a direct
effect of fish predation on the proportion of differ-
ent morphs has little support.
Predation by invertebrate predators: Among the
potential invertebrate predators of C. hirundinella in
our experiment, only Asplanchna was retained in the
RDA model. It has been shown that the presence of
spines in rotifers can reduce their mortality in the
presence of an invertebrate predator such as
Asplanchna (Stemberger and Gilbert 1984), and we
cannot fully discard the hypothesis that the pres-
ence of a third horn could benefit C. hirundinella in
the presence of Asplanchna. However, the contribu-
tion of Asplanchna to the RDA was very small and
only marginally significant.
Neither the univariate nor the multivariate analy-
ses showed any effect of copepods on C. hirundinella
morphology. Moreover, we did not find any negative
relationships between copepods and C. hirundinella
abundance. While we cannot exclude the notion that
this result is due to a time lag between the dynamics
of invertebrate predators and C. hirundinella, it seems
more likely that this result is partly due to the low
palatability of C. hirundinella for copepods (Karabin
1978, Santer and van den Bosch 1994, Santer 1996).
Despite the uncertainty about the underlying
mechanisms, our results give no support to the idea
that invertebrate predators caused high mortality in
C. hirundinella in our enclosures, as previously
suggested by Hamlaoui et al. (1998). Thus, the
hypothesis that morphological variability in C. hirun-
dinella is related to invertebrate predators is not
supported by our results.
Mechanical interference with grazers: The morpho-
logical variation of C. hirundinella was clearly related
to the biomass of cladocerans in our experiment.
Cladoceran biomass variation was partly confounded
by the presence ⁄ absence of fish, since fish had a
direct effect on cladocerans by reducing their mean
body size and total biomass. However, since perch
had a stronger impact on both average size and bio-
mass of cladocerans than roach, the effects of fish
and cladocerans were not completely confounded
(see Table 2). The separate analyses conducted on
‘‘free-ranging fish’’ and ‘‘fishless’’ enclosures give
strong support to this point.
Although it is unlikely that cladocerans con-
sumed C. hirundinella, it is possible that inter-
actions occurred between the two organisms, as
 INDIRECT EFFECTS ON CERATIUM MORPHOLOGY
suggested by our observations of Ceriodaphnia that
contained C. hirundinella cells in their filtering
chambers. Herbivorous cladocerans can temporar-
ily retain large inedible particles in their branchial
chamber and reject them subsequently using their
postabdominal claw to keep their filtering appara-
tus functioning (Gilbert 1988a, Xie et al. 1998).
This behavior can greatly interfere with large Daph-
nia feeding when C. hirundinella abundance is
high, as directly observed by Xie et al. (1998).
Such direct interaction might cause physical dam-
age to C. hirundinella, as suggested by the large
numbers of Ceratium sp. with broken horns
observed by Cyr and Curtis (1999) in enclosures
dominated by large cladocerans. The average cla-
doceran body length was not very large in our
enclosures, but we have no indication that the
effects differed from small to large individuals. In
fact, analyses conducted on bulk cladoceran sam-
ples or only on larger individuals (>800 lm) gave
similar results (not shown). Thus, the appearance
of a third horn might reduce the vulnerability
of C. hirundinella to mechanical interference by
lowering the probability of being swept into the
filtering chamber of cladocerans. Gilbert (1988b)
first suggested the existence of a causal relation-
ship between morphological variation and mechan-
ical interference for Daphnia–rotifer interactions.
Gilbert (1988b) showed that rotifers swimming
close to the carapace of a feeding Daphnia may be
caught by the inhalant current, swept into its bran-
chial chamber and rapidly rejected by the postab-
domen. This phenomenon can cause mortality in
newly hatched rotifers (Gilbert and MacIsaac
1989), with negative effects on the populations of
susceptible species (Gilbert 1988a). Some rotifers
have been shown to counteract this negative effect
of cladocerans by producing offspring with mor-
phological defenses such as spines (Gilbert and
MacIsaac 1989).
In conclusion, after confirming that the occur-
rence of planktivorous fish was strongly associated
with the morphological variation found in C. hirun-
dinella, we can confidently exclude fish predation,
turbidity, and invertebrate predation as direct causal
factors behind this phenomenon (Fig. 4). However,
our results suggest that the third horn in C. hirundi-
nella might be viewed as a protection against
mechanical interference, similar to what has been
demonstrated in the Daphnia–rotifer interaction,
even though cladocerans cannot eat C. hirundinella.
While laboratory experiments are now needed to
test this hypothesis and to understand the role of
other possible pathways, our experiment allowed us
to clearly exclude some factors and to narrow the
search for potential mechanisms behind this phe-
nomenon.
Gérard Lacroix benefited from the ANR French National Pro-
gram (BioFun project). The authors thank Johan Lövgren,
39
Lennart Persson, and Frank Johansson for constructive com-
ments on the results. We also thank Laure Devine, Eric
Pattee, and Sergio Perla for linguistic improvements and con-
structive comments on a previous version of this manuscript.
AFNOR 1990. Eaux – me´thodes d’essais. Association Française de
Normalisation, Paris, 736 pp.
Bertolo, A., Lacroix, G. & Lescher-Moutoué, F. 1999a. Scaling food
chains in aquatic mesocosms: do the effects of depth override
the effects of planktivory? Oecologia 121:55–65.
Bertolo, A., Lacroix, G., Lescher-Moutoué, F. & Cardinal-Legrand,
C. 2000. Plankton dynamics in planktivorous- and piscivorous-
dominated mesocosms. Arch. Hydrobiol. 147:327–49.
Bertolo, A., Lacroix, G., Lescher-Moutoué, F. & Sala, S. 1999b.
Effects of physical refuges on fish–plankton interactions.
Freshw. Biol. 41:795–808.
Cyr, H. & Curtis, J. M. 1999. Zooplankton community size structure
and taxonomic composition affects size-selective grazing in
natural communities. Oecologia 118:306–15.
Gilbert, J. J. 1988a. Susceptibilities of ten rotifer species to inter-
ference from Daphnia pulex. Ecology 69:1826–38.
Gilbert, J. J. 1988b. Suppression of rotifer populations by Daphnia, a
review of the evidence, the mechanisms, and the effects on
zooplankton community structure. Limnol. Oceanogr. 33:1286–
303.
Gilbert, J. J. & MacIsaac, H. J. 1989. The susceptibility of Keratella
cochlearis to interference from small cladocerans. Freshw. Biol.
22:333–9.
Hamlaoui, S., Couté, A., Lacroix, G. & Lescher-Moutoué, F. 1998.
Nutrient and fish effects on the morphology of the
dinoflagellate Ceratium hirundinella. C. R. Acad. Sci. Paris
321:39–45.
Hjelm, J. & Johansson, F. 2003. Temporal variation in feeding
morphology and size-structured population dynamics in fishes.
Proc. R. Soc. Lond. Ser. B 270:1407–12.
Huber-Pestalozzi, G. 1950. Das Phytoplankton des Süsswassers: Sys-
tematik und Biologie. Cryptophyceen, Chloromonadinen, Peridineen
Band 16 Teil 3. E. Schweizerbart’sche Verlagsbuchhandlung,
Stuttgart, Germany.
Hutchinson, G. E. 1957. A Treatise on Limnology, 1st ed. Chapman
and Hall, London, 1015 pp.
Karabin, A. 1978. The pressure of pelagic predators of the genus
Mesocyclops (Copepoda, Crustacea) on small zooplankton. Ekol.
Pol. 26:241–57.
Kim, B. H., Han, M.-S. & Takamura, N. 2003. Effects of fish intro-
duction on the length of the tail of cryptomonads in meso-
cosm experiments. Oecologia 136:73–9.
Legendre, P. & Gallagher, E. D. 2001. Ecologically meaningful
transformations for ordination of species data. Oecologia
129:271–80.
Lürling, M. 2003. The effect of substances from different zoo-
plankton species and fish on the induction of defensive mor-
phology in the green alga Scenedesmus obliquus. J. Plankton Res.
25:979–89.
Nunn, A. D., Harvey, J. P. & Cowx, I. G. 2007. The food and feeding
relationships of larval and 0 + year juvenile fishes in lowland
rivers and connected waterbodies. II. Prey selection and the
influence of gape. J. Fish Biol. 70:743–57.
Popovský, J. & Pfiester, L. A. 1990. Süßwasserflora von Mitteleuropa:
Dinopyceae (Dinoflagellida). Gustav Fisher Verlag, Jena,
Germany, 272 pp.
Quinn, G. & Keough, M. 2002. Experimental Design and Data Analysis
for Biologists. Cambridge University Press, Cambridge, UK, 556
pp.
Santer, B. 1996. Nutritional suitability of the dinoflagellate Ceratium
furcoides for four copepod species. J. Plankton Res. 18:323–33.
Santer, B. & van den Bosch, F. 1994. Herbivorous nutrition of
Cyclops vicinus, the effect of a pure algal diet on feeding,
development, reproduction and life cycle. J. Plankton Res.
16:171–95.
Sommer, U., Sommer, F., Santer, B., Zöllner, E., Jürgens, K.,
Jamieson, C., Boersma, M. & Gocke, K. 2003. Daphnia versus
40 ANDREA BERTOLO ET AL.

copepod impact on summer phytoplankton: functional
compensation at both trophic levels. Oecologia 135:639–47.
Stelzer, C. 1998. Feeding behaviour of the rotifer Ascomorpha ovalis:
functional response, handling time and exploitation of indi-
vidual Ceratium. J. Plankton Res. 20:1131–44.
Stemberger, R. S. & Gilbert, J. J. 1984. Spine development in the
rotifer Keratella cochlearis: induction by cyclopid copepods and
Asplanchna. Freshw. Biol. 14:639–47.
ter Braak, C. J. F. & Smilauer, P. 2002. CANOCO Reference Manual
and CanoDraw for Windows User’s Guide, Software for Canonical
Community Ordination (Version 4.5). Centre for Biometry
Wageningen, Wageningen, the Netherlands, 500 pp.
Xie, P., Iwakuma, T. & Fujii, K. 1998. Changes in the structure of a
zooplankton community during a Ceratium (dinoflagellate)
bloom in a eutrophic fishless pond. J. Plankton Res. 20:1663–
78.