Effects of trophic status on microcystin production and the dominance of cyanobacteria in the phytoplankton assemblage of Mediterranean reservoirs

The aim of our study was to evaluate the abundance of cyanobacteria and microcystins in four Sardinian reservoirs (Italy) characterised by different trophic status to define a reference picture for future changes. Increasing levels of eutrophication and the abundance of cyanobacteria are expected to occur due to climate change, especially in the southern Mediterranean. Consequently, an in-depth study of the occurrence of harmful cyanobacteria is important to develop appropriate management strategies for water resources at a local scale. Monthly samples were collected at one station in each reservoir over an 18-month period. The Analysis of similarity indicated that cyanobacterial abundance and species composition differed significantly among the reservoirs. The Redundancy analysis highlighted their relationship to trophic, hydrological and seasonal patterns. Spearman’s analysis indicated that there were significant correlations among the most important species (Planktothrix agardhii–rubescens group, Aphanizomenon flos-aquae and Dolichospermum planctonicum), nutrients and microcystins. We highlighted that the species composition during periods of maximum microcystin concentrations differed from those typically reported for other Mediterranean sites. We found new potential microcystin producers (Aphanizomenon klebahnii, Dolichospermum macrosporum and Dolichospermum viguieri), which emphasised the high diversity of cyanobacteria in the Mediterranean area and the need for detailed research at the local scale.


Environmental variables and phytoplankton.
Monthly rainfall values, air temperature and wind intensity, as recorded by the ENAS's meteo-climatic station located close to the dams, were considered during the period from January 2010 to December 2011 to further characterise the reservoirs. The data for TOR were derived from Sardegna Clima Onlus data network (http://www.sardegna-clima.it), considering the meteo-climatic station closest to the reservoir (Tonara).
Transparency was assessed based on Secchi disk (SD) measurements. Temperature (T), conductivity (CD), pH, and dissolved oxygen (DO) were measured in the field using a multi-parametric probe (Hydrolab Datasonde 5). The values recorded in the field were replicated in the laboratory for pH (Orion Research Model 960, Beverly, MA, USA) and conductivity (Analytical Control 120).
Phytoplankton samples (100 ml) were fixed in the field using Lugol's solution and analysed using Utermöhl's technique 28 . Cell abundance was determined microscopically from subsamples (5-10 ml) of the fixed samples, using an inverted microscope (Zeiss, Axiovert 10, Oberkochen, Germany) at 200 × and 400 × magnifications, based on cell counts from an appropriate number of fields. Species were identified from samples of live and fixed  30,31 , and Suda et al. 32 . The taxon bio-volume was determined by multiplying the cell abundance of each taxon by the cell volume, which was obtained by geometrical approximations from the measurement of at least 30 specimens, following the method of Sun and Liu 33 . The bio-volume was converted to biomass, based on the assumption that 1 mm 3 = 1 mg fresh weight biomass 34 . The environmental and nutrients data were averaged for the entire water column. The Chl a and phytoplankton data were averaged for the euphotic zone (Zeu = 2.5 times the Secchi disk depth 35 ). The Zeu/Zmix ratio (where Zmix is the mixing depth) was used as a measure of light availability in the mixed layer 36 . The relative water column stability (RWCS) was calculated following Naselli-Flores 37 .
Following Reynolds 38 , we related the dominant and co-dominant phytoplankton species (in terms of biomass) to C-S-R (competitor, C; stress-tolerant, S; ruderal, R) life strategies 38,39 to interpret the phytoplankton responses, in terms of growth strategies, to various environmental conditions at both temporal (seasonal) and spatial (site) scales.
The trophic status of the four reservoirs was assessed based on Vollenweider-OECD classification criteria 40 .
Toxin determination. MCs and nodularins were detected in water samples by an enzyme-linked immunosorbent assay (ELISA), following Fischer et al. 41 . After a freeze-thaw cycle and ultrasonic treatment (ELMA S 10, Elmasonic), the samples were analysed using an ELISA kit (Microcystins ADDA-ELISA Microtiter Plate Abraxis LLC, Warminster, PA), according to the manufacturer's protocol. Absorbance was read at a single wave length of 450 nm using a Miniphotometer (model 6 + , Metertech Inc.). The results were expressed as MC-LR equivalents, as indicated by the manufacturer, and were reported in the text as MCs. Samples were considered positive when the MCs concentration was higher than the lowest detection limit (0.10 μ g l -1 ).

Data analyses.
Non-metric multidimensional scaling analyses (nMDS) of species bio-volumes were conducted to assess the differences among sites. A similarity matrix was obtained using a dataset comprising all the cyanobacterial species. The similarity matrices were based on the Bray-Curtis similarity index. The significance of spatial differences was assessed using a one-way analysis of similarities test (ANOSIM) 42 , and probability percentages of less than 3% were considered significant. The major species contributing to the spatial differences among reservoirs and the similarity among samples at each reservoir were investigated using a similarity percentage analysis (SIMPER) 42 . This analysis broke down the percentage contribution of each species, allowing identification of the species that were most important in creating the observed pattern of difference. A data matrix was constructed considering all the cyanobacterial species abundances in all the samples for each reservoir. The Bray-Curtis similarity measure was implicit to SIMPER. Ordinations were performed using CANOCO 43 . A detrended correspondence analysis (DCA) was first conducted on potentially toxic cyanobacterial species (selected via SIMPER) and MCs concentrations and indicated a linear distribution (with a gradient length between 2 and 3), which validated the use of a direct linear methodology, such as a redundancy analysis (RDA). For the ordination analysis, the data for abiotic and biological variables were transformed by log 10 (x + 1). RDA was used to examine the relationships between the abiotic variables (T, CD, Zeu/Zmix, RWCS, NH 4 -N, NO 2 -N, NO 3 -N, TN, RP and TP) and potentially toxic cyanobacterial species and to select the variables that best described the seasonal distribution of potentially toxic cyanobacterial species. The significance of environmental variables in explaining the variance of potentially toxic cyanobacterial species in the RDA was tested using Monte Carlo simulations with 499 permutations.
Non-parametric correlation (Spearman) analyses were used to test the relationships between species selected via SIMPER and environmental variables (SD, T, pH, CD, Alk, DO, Zeu/Zmix and RWCS), nutrients (NH 4 -N, NO 2 -N, NO 3 -N, TN, RP and TP), and Chl a and MCs concentrations. For these analyses, we used the data for the cell abundance of potentially toxic cyanobacterial species at 0.5 m, which was the same depth used for the MCs analyses. A Spearman analysis was conducted using XLSTAT 10.10.

Results
Environmental conditions and nutrients. During the study period, the four reservoirs were characterised by different environmental conditions and nutrient concentrations ( Table 2). TP and TN clearly increased from TOR to SCN, PAT and BID (Fig. 2). Similarly, there were gradients among the reservoirs for SD transparency, water temperature and Zeu values (Table 2, Fig. 2). There was a stable thermal stratification from June to October in PAT (Zmix: 9.  Table 2). The Zeu/Zmix ratio was lowest in the winter-spring months (minimum of 0.03 in PAT in April, 0.04 in BID in December, 0.08 in SCN in February and 0.21 in TOR in April) and highest in summer (maximum of 1.4 in BID in July, 1.5 in PAT in July, 2.5 in SCN in August and 3.0 in TOR in June).
The air temperature increased from SCN to TOR, PAT and BID. The wind intensity increased from TOR to BID, PAT and SCN, and rainfall increased from SCN to PAT, BID and TOR (Table 1). Wind intensity was less variable in BID than in the other reservoirs. Rainfall showed a very similar seasonal pattern in all of the reservoirs, with the maximum in October-December and the minimum in May and August (Table 1, Fig. 3).
According to the Vollenweider-OECD classification criteria 40 , the previously assessed trophic status were confirmed for all the reservoirs (Table 1).

Phytoplankton.
Chl a, cell abundance and biomass values increased from TOR to SCN, PAT and BID (      In SCN and TOR, the phytoplankton biomass was dominated by Dinophyceae (dominant species: Gymnodinium uberrimum (G.J.Allman) Kofoid and Swezy) and by Cryptophyceae in TOR (dominant species: Plagioselmis lacustris (Pascher and Ruttner) P. Javornický) (Fig. 4). S-strategists (i.e., stress tolerant species), which prefer environments with low nutrient concentrations but high energy conditions, were dominant during stratification periods (high Zeu/Zmix) and prevailed in the reservoirs that had lower trophic status (SCN and TOR; e.g., Gymnodinium). R-strategist species (i.e., ruderal species), which are adapted to maximising suspension opportunities and low irradiance values, characterised the periods with low Zeu/Zmix values (mixing periods) in PAT (e.g., Planktothrix spp.) and those with high Zeu/Zmix conditions (stratification periods) in BID (e.g., Aulacoseira), preferring reservoirs that had higher trophic status. C-strategists (i.e., competitive species), which are adapted to environments with low disturbance levels and saturated by light and nutrients, were particularly present during mixing periods in both hypereutrophic (BID; e.g., Cyclotella and Stephanodiscus species) and mesotrophic conditions (TOR; e.g., Plagioselmis lacustris). Among the cyanobacteria (Table 3), S-strategist and C-strategist species were the most numerous, especially under hypereutrophic and eutrophic status.
Cyanobacterial species composition and microcystins. Cyanobacterial species were more numerous in BID (32) and PAT (31) than in SCN (10) and TOR (8) ( Table 3). The three dominant species, in terms of cell abundance, in hierarchical order were Aph. flos-aquae, Cyanocatena imperfecta (Cronberg & Weibull) Joosten and Aphanocapsa spp. in BID; Planktothrix agardhii-rubescens group, Aphanocapsa spp. and Aph. flos-aquae in PAT; Anathece sp., Dolichospermum viguieri (Denis and Frémy) Wacklin, Hoffmann and Komá rek (= Anabaena viguieri Denis and Fremy) and Planktothrix agardhii-rubescens group in SCN; and Anathece sp., Aphanocapsa spp. and Dolichospermum spp. in TOR (Fig. 5). The ANOSIM revealed significant differences in the cyanobacterial assemblages among the four reservoirs (ANOSIM Global R = 0.55, p < 0.001). The nMDS analyses showed that the data clusters were more compact for PAT and TOR (Fig. 6) and, therefore, that the differences among samplings were minimal in these reservoirs. These findings were supported by the SIMPER results, which indicated that the high level of data similarity in samples from PAT (33.69%) was attributable to the P. agardhii-rubescens group (75.57% contribution). The greatest differences in the cyanobacterial assemblages were between PAT and SCN (99.50%), between PAT and TOR (99.35%), and between BID and SCN (99.30%). There were smaller but significant differences between SCN and TOR (97.81%) and between BID and PAT (94.02%). P. agardhii-rubescens group, Aph. flos-aquae, Aphanizomenon sp. and Dolichospermum planctonicum (Brunnthaler) Wacklin, Hoffmann and Komá rek (= Anabaena planctonica Brunnth) were the species that were mainly responsible for the differences in cyanobacterial assemblages (contribution > 10%) between PAT and BID and between these two reservoirs and SCN and TOR. The species mainly responsible for the differences in cyanobacterial assemblages (contribution > 10%) between SCN and TOR were  We detected potentially toxic cyanobacterial species in all of the reservoirs ( Table 3). The potentially toxic species identified in this study included the first records for Sardinia for: Dolichospermum macrosporum (Klebahn) Wacklin, Hoffmann and Komá rek (= Anabaena macrospora Klebahn) in BID and PAT; D. viguieri in BID, PAT and SCN; and Aphanizomenon klebahnii (Elenkin) Pechar and Kalina in BID (Fig. 7).
MCs were detected in 71% of the analysed samples (n = 62), including 43% of samples from TOR (n = 14), 63% of samples from SCN (n = 16), 75% of samples from BID (n = 16), and 100% of samples from PAT (n = 16). A concentration > 1 μ g l -1 MC-LR was found in almost half of the samples but was found only once in TOR and never in SCN (Fig. 5).
The highest concentrations of MCs were in the summer-autumn period in BID, SCN, and TOR. The highest MC concentrations coincided with high cell abundances at the superficial depth of Aphanocapsa sp. and Aph. flos-aquae in BID, D. viguieri in SCN, and Dolichospermum sp. (Anabaena sp.) in TOR. In contrast, the maximum MCs concentrations were in PAT in winter-spring; these were the highest concentrations detected in this study. In PAT, the maximum MCs concentrations in winter-spring coincided with the highest cell abundances of cyanobacteria from October to April (from January to April, cell abundance of > 181 × 10 6 cells l -1 , and MCs concentration of > 4 μ g l -1 ), due to the P. agardhii-rubescens group affirmation. In particular, in samples from December 2010 to April 2011, the dominant species was P. rubescens (De Candolle ex Gomont) Anagnostidis and Komárek (= Oscillatoria rubescens De Candolle ex Gomont) (Fig. 7).

Discussion
Trophic status, environmental conditions and phytoplankton. Based on the Vollenweider-OECD classification criteria 40 , there was a trophic gradient among the four reservoirs. These data were consistent with the results of previous studies 16,17 and supported our choice of these reservoirs as prototypical examples of reservoirs with different trophic status for inclusion in our study.
Nutrient enrichment of water bodies (eutrophication) is considered the main driver of cyanobacterial abundance 8,44 and persistence 23,45,46 . Therefore, we expected that the occurrence of cyanobacteria and their toxic blooms would increase along the trophic gradient of the reservoirs. We found that cyanobacterial abundance differed significantly among the reservoirs (ANOSIM results), with greater abundance in the eutrophic and hypereutrophic PAT and BID than in the oligo-and meso-eutrophic SCN and TOR. The RDA clearly highlighted this relationship, and Spearman's analysis indicated that there were significant correlations among the most abundant species in eutrophic reservoirs (i.e., P. agardhii-rubescens group, Aph. flos-aquae and D. planctonicum), nutrients, and MCs. The RDA also revealed seasonal patterns. These findings were consistent with the ecological preferences of many potentially toxic cyanobacterial species, such as Aph. flos-aquae and D. planctonicum, which prefer high summer temperatures and high water stability 16 . By contrast, the ecological preferences (low irradiance and low water  stability) of the P. agardhii-rubescens group agreed with its autumn-winter affirmation in PAT. Although there was a temperature gradient among the four studied reservoirs ( Table 2, Fig. 2), their monomictic character was evident from the stable thermal stratification throughout the summer-autumn period. In general, during thermal stratification, the hydrology of Mediterranean reservoirs favours nutrient accumulation in the deeper water layers. The nutrient levels can increase because of nutrient release from sediment under anoxic conditions, and long periods of stratification can increase hypolimnetic oxygen depletion 47 . Operational procedures and water level fluctuations in summer months can cause the breakdown of the thermal stratification and the re-distribution of nutrients during the period that is most favourable for phytoplankton growth 48 . Moreover, the water turbulence generated during the emptying phase, along with the disruption of stratification, can increase the amount of particulate material in the water column because of re-suspension from the bottom 49 . Based on our data, these effects appear to be less pronounced in these Sardinian reservoirs because they are generally deeper and more interconnected, thus ensuring the persistence of stratification until autumn 16,18,25 . Similar scenarios for other deep Mediterranean reservoirs were reported in other studies. For example, seasonal variations in thermal stratification were maintained during selective water withdrawal 50 and during refilling of the reservoir via a water pump located on the river 51 or via the natural flow of a river tributary 36 .
Long periods of stable stratification are competitively advantageous for gas-vacuolated cyanobacterial species (e.g., Microcystis, Dolichospermum, and Aphanizomenon), which can regulate their position in the water column (buoyancy) to optimise the use of resources (nutrients and light) 8 . These genera were reported to be the main CyanoHAB producers in the Mediterranean in the summer-autumn period 13 . In our study, a similar case was observed in BID and SCN, where Aphanizomenon and Dolichospermum (Anabaena) were important genera. Aphanizomenon and Dolichospermum (Anabaena) are considered S-strategists, preferring high hydrological stability and high Zeu/Zmix 38,52 . By contrast, long-lasting stratification with consequent high levels of nutrient enrichment of the hypolimnion, followed by mixing of the entire water column in late autumn, can significantly affect the cyanobacterial species composition, thus favouring, for example, Planktothrix blooms 8 . This was the case for PAT, which was subjected to a longer stratification period and the extensive growth of Planktothrix species from late autumn to spring during mixing. Planktothrix species are considered R-strategists because they are adapted to maximising suspension opportunities and low irradiance values 38 .
Species composition of Cyanobacteria. The occurrence of more than 20 species of potentially toxic cyanobacteria followed the trophic gradient in PAT and BID, whereas in SCN and TOR there were 10 or fewer species. We identified cyanobacterial species that were previously reported in Sardinian reservoirs 17,18 and from other Mediterranean sites 13,52 . However, we also found species that were not previously reported in these areas. These species included D. macrosporum (Klebahn) Wacklin, Hoffmann and Komá rek (Fig. 4), Aph. klebahnii (Elenkin) Pechar and Kalina, and D. viguieri (Fig. 4). D. macrosporum (found in BID and PAT) was reported in Britain 53 , France 54 , Romania 55 , Germany 56 , Spain 57 and the Czech Republic 58 , and it is known to be toxic 57 . Aph. klebahnii (found in BID) was reportedly distributed in temperate zones 59 , the Baltic Sea 60 , and Lithuania 61 , and it is thought to be potentially harmful because it produces MCs 62 . D. viguieri (found in BID, PAT and SCN) was reported in Britain 53 , the Czech Republic 63 , Greece 64 , and Russia 65 , and it is a potential MCs producer 66 .
Aphanocapsa sp., Aph. flos-aquae, D. viguieri, Dolichospermum sp. (Anabaena sp.), and the P. agardhii-rubescens group, especially P. rubescens, were the most important in determining the maximum of the cyanobacteria abundance in the studied reservoirs. This result suggested that there was a marked reduction in the occurrence and importance of Microcystis compared with previous scenarios in the same reservoirs. M. aeruginosa (Kűtzing sensu Teiling) and M. flos-aquae (Wittrock) Kirchner were the dominant cyanobacteria in the summer-autumn periods in BID 17 , and Microcystis spp. were dominant in the summers of 1988, 1997, and 1999 in PAT 67 . Planktothrix spp. dominated the winter-spring months between 2004 and 2005 in TOR 20 but were rarely found in this study. Dolichospermum planctonicum and D. ellipsoides (Bolochoncev) (= Anabaena elliptica Lemmermann) were reported as dominant in SCN in the summer-autumn months of 1993, following the strong dominance of Aphanocapsa and Anathece of 1991 and 1992 68 . In the present study, we detected significant quantities of these members of the Chroococcales and detected D. viguieri for the first time.
Cyanobacteria and toxicity. We found a progressive increase in the occurrence of cyanobacterial abundance and MCs in response to increasing trophic status in these four Sardinian reservoirs. Nutrient over enrichment of waters by human pressure, such as wastewaters from urban centres, agricultural, and industrial activities, has promoted the growth of cyanobacteria as harmful algal blooms 10 . In most cases, the delimitation of the factors controlling and promoting the development of cyanotoxins was not clear and was very difficult to address 69 . Focusing on MCs, several studies revealed a huge variability in the concentrations. Salmaso et al. 69 indicated that most of this variability was due to differences in the cyanobacterial biomass in lakes of different trophic status. The authors also highlighted the importance of the fraction of toxic and non-toxic genotypes within cyanobacterial species and physiological variation in toxin quota (MC concentration per cell or unit biomass) in toxic populations. In approximately 50% of the samples analysed in our study, the concentration of MCs exceeded the MC-LR drinking water guideline (> 1 μ g l -1 ) 70 , which indicated that the waters of the hypereutrophic BID and, especially, the eutrophic PAT posed a significant threat to human health. Surprisingly, the WHO limit was exceeded once in TOR, and MCs were detected on other occasions, including in SCN. Messineo et al. 20 reported that MCs were detected in TOR in 2004 and 2005, corresponding to abundant Planktothrix, but the concentrations were always low (< 1 μ g l -1 ). Overall, the proportion of positive samples was similar to that reported in a number of studies, including those conducted in other countries 13,64 . Our data further emphasised the need for the continuous monitoring of cyanobacterial toxicity, especially in areas where resources are limited, such as in Sardinia and the Mediterranean basin in general.
In our study, the maximum MCs concentrations were detected in summer-autumn, when the reservoirs were stratified and the temperature was 20-30 °C, except in PAT. This seasonality was consistent with that observed in other studies on Mediterranean reservoirs 13,71 . However, Cook et al. 64 indicated that cyanobacterial blooms may start before summer and persist until after autumn. Cook et al. 64 and Carrasco et al. 13 also reported that M. aeruginosa was the most widespread and most frequently occurring toxic species in the entire Mediterranean region, thus confirming the findings of Abdel-Rahman et al. 72 and Oudra et al. 73 . A toxic strain of M. aeruginosa was isolated from a reservoir in northern Sardinia, Lake Liscia, and was associated with a fish-kill event 19 . In our study, M. aeruginosa was never dominant during periods of peak MCs concentrations. Rather, these periods coincided with the occurrence of high numbers of Chroococcales in the genus Aphanocapsa (BID); N 2 -fixing Nostocales, including Aph. flos-aquae (BID) and D. viguieri (SCN) and Dolichospermum sp. (Anabaena sp.) (TOR); and Oscillatoriales, including members of the P. agardhii-rubescens group (PAT). Aph. flos-aquae dominated in the more eutrophic reservoirs (BID and PAT). Spearman's correlation analysis and RDA indicated that the occurrence of Aph. flos-aquae was significantly correlated with the main nutrients (RP, TP, NH 4 -N, and TN) and temperature, which was consistent with its preference for summer months. Aph. flos-aquae was linked to saxitoxin production in the United States. Aphanizomenon spp. was linked to saxitoxin production in Greece 74 , anatoxin-a production in Finland and Germany 75 and cylindrospermopsin production in Poland 76 and Germany 77 . However, the production of MCs by Aphanizomenon species has not been conclusively demonstrated 71,78 . In our case, although the period of peak MCs partly coincided with high cell abundances of Aph. flos-aquae in BID and PAT, it was likely that other cyanobacterial species present at the same time (i.e., Microcystis spp., D. planctonicum and Aphanocapsa spp.) were responsible for the high MCs concentrations. In contrast, the toxicity of various Dolichospermum (Anabaena) species, including D. vigueri, was well documented 75,79,80 . A strain of D. planctonicum (= A. planctonica) from Mulargia Lake (south-central Sardinia) was shown to produce unidentified MC-like peptides 20 , and this species dominated in Sardinian reservoirs whose waters contained high concentrations of extracellular MCs 20 . Finally, because large populations of Aphanocapsa species coincided with high MCs concentrations in BID and PAT, further studies are required to evaluate their toxicity and ecology in the Mediterranean basin. To date, the ability of Aphanocapsa to produce MCs was only demonstrated for Aphanocapsa cumulus 81 from Botswana, which is more widely distributed in tropical regions 82 .
The MCs maxima occurred in winter-spring in PAT during a bloom of the P. agardhii-rubescens group, which confirmed the occurrence of harmful strains of this group in Mediterranean reservoirs. This result also illustrated that the ecological requirements of these species differed from those of the other species in this study. In general, Planktothrix show strong growth in relatively low-nutrient conditions, potentially enabling them to inhabit waters with a wide range of trophic status 69 . Since the mid-1980s, these species have been reported from a number of Mediterranean sites, including natural lakes 83 , reservoirs 20,84,85 , and lagoons 86 . The abundance of Planktothrix increased in the more eutrophic lakes south of the Alps 87 . In Lake Garda, the appearance of mesotrophic and eutrophic species (i.e., P. rubescens) after the 1970s and during the 1980s illustrated the strong effect of eutrophication on the phytoplankton community 88 . A similar trend was observed in Lake Maggiore over a long period 89 . Among the species in this group, P. rubescens is one of the most widespread toxin-producing cyanobacteria in European lakes 83,90 . It is thought to be the species that produces the highest concentration of MCs per cell among all known cyanobacteria 91 . The red-pigmented P. rubescens was reported to be the most representative cyanobacterium in the subalpine region, and it is common in deep and well-stratified lakes 92,93 . In Sardinia, P. rubescens caused the extremely dangerous cyanotoxin events that occurred in the winter months in the Mulargia and Flumendosa Lakes in 1985 20 . By contrast, P. agardhii (Gomont) Anagnostidis and Komàrek, another species in the P. agardhii-rubescens group, is most frequently observed in shallow and eutrophic lakes and is frequently found in the northern hemisphere 84 . Bonilla et al. 94 reported that P. agardhii (Gomont) Anagnostidis and Komárek is common in eutrophic freshwaters, based on analysis of a large database (940 samples) covering different climatic regions and the Northern and Southern hemispheres. Toxic strains of this species were also isolated from several European lakes 95 .
The most abundant cyanobacterium in this study, P. rubescens, was reported to produce concentrated metalimnetic populations during thermal stratification in summer and autumn, when sufficient light penetrated into these water layers 93,96 . The presence of P. rubescens becomes evident when the trichomes move towards the surface during mixing in winter or when the water turbidity increases. The vertical distribution of the P. agardhii-rubescens group in PAT was consistent with this pattern (data not shown). In Sardinian reservoirs, transparency minima were linked both to summer phytoplankton growth and winter sediment runoff from the catchment 97 , accentuated by the strong seasonality of rainfall. Consistent with these results, the Spearman's correlation analysis and RDA in this study indicated that the P. agardhii-rubescens group was negatively correlated with Secchi disk transparency. In any case, this relationship may depend, at least partly, on the increased turbidity resulting from the abundance of the P. agardhii-rubescens group in the mixing period. Excessive growth of this species produces characteristic reddish water discoloration, which generates alarm among nearby residents and lake managers. Our data suggest that this alarm is warranted because of the high toxin concentrations in PAT during the periods of maximum cell abundance of this species. On these occasions, the WHO MC limit was substantially exceeded for long periods (Fig. 5).

Conclusions
Our study complemented and updated what was already known for the study areas [16][17][18]20,25,67,68,98,99 and enabled us to support the findings of other similar case studies in other geographical areas 97,100,101 . Our study provided important evidence for the strong relationships among trophic status, cyanobacterial abundance, and MCs concentrations in Sardinian reservoirs, indicating a key role of nutrients in determining cyanobacteria affirmation. Moreover, we confirmed that the temperature and Zeu/Zmix ratio were important factors in cyanobacteria development 10,12,94,102,103 . However, the water management of reservoirs, linked to the different needs and uses, further increases the ecosystem complexity and the difficulty in understanding the multiple relationships between biotic and abiotic components.
We also highlighted that the species composition during periods of maximum MCs concentrations in the four reservoirs differed from those typically reported for other Mediterranean sites 13,64 and that the P. agardhii-rubescens group was dominant during the period in which the MCs concentrations substantially exceeded the WHO limit. Our results included the first reports of some potentially toxic species in reservoirs of the Mediterranean basin, emphasising the importance of long-term studies for the early detection of community changes. This is especially important when the changes involve harmful species. Phytoplankton responses to nutrient variations cannot be separated from responses to other, larger environmental changes occurring at the same time, such as global climate change. In particular, increasing levels of eutrophication and the abundance of cyanobacteria are expected due to climate change, especially in the southern Mediterranean and islands. Cyanobacteria are also expected to be favoured by higher temperatures 10,103 , especially at lower levels of nutrient concentrations 12 . In any case, the progressive enlargement of the geographical distribution of harmful species both in Mediterranean area and globally [104][105][106] highlights the need for further detailed research on their ecology, toxicology, and genetics at the local scale.