Multiple interacting environmental drivers reduce the impact of solar UVR on primary productivity in Mediterranean lakes

Increases in rainfall, continental runoff, and atmospheric dust deposition are reducing water transparency in lakes worldwide (i.e. higher attenuation Kd). Also, ongoing alterations in multiple environmental drivers due to global change are unpredictably impacting phytoplankton responses and lakes functioning. Although both issues demand urgent research, it remains untested how the interplay between Kd and multiple interacting drivers affect primary productivity (Pc). We manipulated four environmental drivers in an in situ experiment—quality of solar ultraviolet radiation (UVR), nutrient concentration (Nut), CO2 partial pressure (CO2), and light regime (Mix)—to determine how the Pc of nine freshwater phytoplankton communities, found along a Kd gradient in Mediterranean ecosystems, changed as the number of interacting drivers increased. Our findings indicated that UVR was the dominant driver, its effect being between 3–60 times stronger, on average, than that of any other driver tested. Also, UVR had the largest difference in driver magnitude of all the treatments tested. A future UVR × CO2 × Mix × Nut scenario exerted a more inhibitory effect on Pc as the water column became darker. However, the magnitude of this synergistic effect was 40–60% lower than that exerted by double and triple interactions and by UVR acting independently. These results illustrate that although future global-change conditions could reduce Pc in Mediterranean lakes, multiple interacting drivers can temper the impact of a severely detrimental driver (i.e. UVR), particularly as the water column darkens.


Results
Water transparency of Mediterranean lakes in worldwide comparison. We used the attenuation of PAR (400-700 nm) as a measure of transparency of the water column; we found low median values of Kd PAR in aquatic ecosystems worldwide (~ 0.50 m −1 ), implying generally high transparency at these wavelengths of solar radiation (Fig. 1). Median Kd PAR values were 0.49 m −1 in boreal/polar, to 0.65 m −1 in temperate, and 0.54 m −1 in tropical lakes. No significant differences in Kd PAR were found among climatic areas due to its high variability (LSD post hoc test, p > 0.20; n = 421). In our experiments with phytoplankton communities from nine Mediterranean lakes, the water had a Kd PAR gradient of 0.18-0.90 m −1 , this falling within the range found on a global scale (Fig. 1).
Individual and interactive effects of UVR, CO 2 , Mix and Nut on P c . The P c rates ranged between 0.02 and 0.03 and 0.80 h −1 (Fig. S1). In six of the nine lakes sampled, all drivers tested decreased the P c (ranging − 0.05 and − 1.80) although Nut and Mix proved less inhibitory [ln response ratio (lnRR) < 0.52 in all cases] than did UVR or CO 2 (Fig. 3A-D).
UVR was the dominant driver because the magnitude of its effect was higher (ca. 3-60-fold in average) than that exerted individually by all the other drivers tested. Also, UVR had the largest difference in driver magnitude of all treatments. Double and triple interactions among drivers revealed two response patterns: (1) a weaker inhibitory effect exerted by each single driver on P c , particularly UVR and CO 2 ; and (2) an antagonistic effect on P c in ~ 50% of the double and triple interactions tested ( Fig. 3E-J). The UVR × CO 2 × Nut × Mix interaction synergistically reduced P c (i.e. Río Seco Superior; Fig. 3K). However, this negative effect (minimum lnRRinteractive values < − 0.7) was some 40-60% lower than exerted in 2-level (i.e. minimum lnRR values of ~ − 1.5), and . Finally, we found no significant effect for the TN:TP ratios, in situ temperature, or mean solar irradiance during exposure on the P c response to UVR, CO 2 , Mix, and Nut, and the interaction of any of these factors (Table S2).   www.nature.com/scientificreports/   Table S3). Altogether, the mean lnRR single was − 0.23, while lnRR double and lnRR triple were 0.04 and 0.03, respectively, and lnRR interactive was 0.19. This signifies that the inhibitory effect (mostly by UVR) decreased as the number of interacting drivers increased.

Discussion
Our work evidences that water transparency (i.e. estimated as Kd PAR ) can be a key predictor of the effects of multiple environmental drivers may exert on P c in Mediterranean lakes. Although our findings do not allow global projections of trends (because they were focused on Mediterranean lakes), here we show that an increasing environmental complexity reduces the magnitude of the individual effects of different global-change drivers, particularly those of UVR, the dominant driver. These findings at the community level extend the proposal by Brennan and Collins 9 who used a model green algal species to state that the greater number of interacting drivers, the more likely that the interaction contains at least one severely detrimental driver, and therefore, that the biotic response to multiple environmental drivers greatly depends on the response to the single dominant driver (i.e. UVR in our case). It bears mentioning that these results represent acute responses of phytoplankton physiology to multiple interacting drivers because our short-term in situ experiments did not let these communities to acclimate/ Figure 4. Natural logarithm response ratios (lnRR) of the single and interactive effects of ultraviolet radiation (UVR), carbon dioxide (CO 2 ), mixing (Mix) and nutrients (Nut) as a function of the photosynthetically active radiation attenuation coefficient (Kd PAR , m −1 ) for the lakes considered in the study. The bars represent the mean of three replicates, and the vertical lines the pooled standard deviation for equal sample size (see "Methods"). The solid and dashed lines represent the fitted polynomial regression lines and the 95% confidence bands, respectively. The values > 0 denote a synergistic effect, and < 0 an antagonistic effect. www.nature.com/scientificreports/ adapt to the predicted environmental conditions. Still, our results are quite realistic for the following reasons: (1) we exposed the communities to predicted future environmental scenarios (RCP 8.5 scenario) 30 , and (2) we worked with natural communities already adapted to different in situ environmental conditions that resembled the median Kd PAR values found in freshwater ecosystems worldwide (Fig. 1). We found that UVR was the driver that exerted the strongest synergistic (and inhibitory) effect on P c , intensifying with the darkening of the water column. These findings are consistent with the results of Helbling et al. 18,23 , who reported a maximum UVR inhibitory effect on P c in turbid rather than clear environments. This response pattern could be explained by the fact that the dominant phytoplankton groups in our lakes (diatoms and dinoflagellates) have lower amounts of photoprotective compounds (e.g. mycosporine-like amino acids [MAAs]) as compared to other phytoplankton groups 31 . Since DOM also acts as a shelter for solar radiation, communities inhabiting these ecosystems have lower natural amounts of MAAs than do organisms adapted to clear (and highly UVR-exposed) environments 28,32 . The lower photoprotective capacity could in turn support the reduced P c reported in our experiments. The potential mechanism underlying this response could involve alterations of the Calvin cycle and RuBisCO regulation 33 . This downregulation mechanism would cause the electron-transport system to accumulate excessive reducing power that could not be dissipated as heat through non-photochemical quenching (NPQ), thus ultimately depressing photosynthesis [e.g. UVR 18 ; fluctuating light 34 ]. Nevertheless, we speculate that the reduced P c could result when a fraction of the C incorporated by photosynthesis is diverted to synthesize ATP and C-rich storage products (e.g. polysaccharides). These molecules are required for energetically costly processes (e.g. photorespiration, nutrient uptake, electron flows) that enable phytoplankton to cope with stressful environmental conditions 35 .
Partially in agreement with our hypothesis, the UVR × CO 2 × Nut × Mix scenario exerted a synergistic effect on P c at Kd PAR > 0.7 m −1 . However, contrary to our expectations, this synergistic effect in darker waters was of lower magnitude under UVR × CO 2 × Mix × Nut than that caused by UVR × Nut, UVR × CO 2 × Nut, UVR × Mix × Nut or by the single UVR effect. The mechanisms underlying this response pattern could be an increased productivity efficiency by natural phytoplankton communities, as a strategy to compensate for reduced photon fluxes in darker environments 36 . It is plausible that the chronic exposure, and consequently the adaptation already under way in communities naturally exposed to high UVR levels, had produced a stress-induced community tolerance 37 . Nevertheless, it should not being forgotten that the inherent greater environmental variability of smaller aquatic ecosystems (i.e. ponds, lakes) also fosters greater potential to cope with the impacts of multiple drivers 10 . As the effects of UVR, CO 2 , Mix, and Nut on phytoplankton vary according to its ecophysiological traits and the ecosystem properties, the way in which the intensity and duration of such drivers will impact aquatic ecosystems in a stressful world remains as a challenge for scientific community.
Within this framework, we suggest that communities from darker environments (e.g. humic lakes, deep epilimnion, estuarine areas) may have a potential competitive advantage when multiple drivers interact. Because ~ 60% of the total solar energy absorbed by phytoplankton in surface clear waters is dissipated as heat (i.e. as NPQ), phytoplankton inhabiting the most illuminated layer of aquatic ecosystems would be operating at about half of their maximal photosynthetic energy-conversion efficiency 38,39 . According to our findings, we could expect a reduction of the maximal photosynthetic efficiency in phytoplankton communities to be lower in darker than in clear surface waters of Mediterranean lakes under the action of multiple drivers. Although each ecosystem has its own particularities, we can rule out the possibility that our results were biased by interference derived from different nutrient ratios, in situ temperatures, or variable mean solar irradiance, as we found no significant effect of these three drivers on P c .

Conclusion
Our research adds to the recent evidence indicating major changes in the structure of the planktonic communities when lakes undergo a light-regime shift towards turbid environments 40 . We propose that low water transparency under multiple interacting drivers may have a synergistic effect on near-surface P c , reducing it by 40%; however, the magnitude of such negative impact would be 40-60% lower than when UVR acts as a single driver. Because our results refer to small spatial and short temporal scales, further studies performed over longer-term scales and at the ecosystem level would enable planners not only to quantify the magnitude of global change with more accuracy but also to design more appropriate management and conservational strategies.

Methods
Literature review: water transparency in lakes. We surveyed the literature from 1960 to 2018 through Scopus using "lake, ultraviolet radiation, attenuation, lake water, dissolved organic carbon, organic matter and freshwater environment" as keywords, along with unpublished data sources by our group, and we found a total of 421 valid estimates related with Kd PAR ("Supplementary dataset"). From this dataset, we calculated the median Kd PAR for temperate, tropical, and boreal/polar lakes. We used Kd PAR as a proxy of the underwater light environment under which phytoplankton is adapted because it is an inherent property of each water mass sampled and consequently does not depend on transient weather conditions such as incident solar radiation. In addition, Kd PAR data are easily available in the literature in comparison with other better descriptors of the underwater light environment such as average irradiance.

Experimental study
Nine lakes from the Sierra Nevada National Park and Lagunas de Ruidera Natural Park were used to establish a gradient of Kd PAR . Lakes from Sierra Nevada National Park are mixed, oligotrophic high-mountain lakes located above the tree line on a siliceous bedrock in a glacial cirque 41 (Table 1), we exposed them to the same experimental manipulation procedure and drivers, as described below. Once in the laboratory, the original water sample collected from each lake (sampling- Fig. 5) was divided into 12 2-L polyethylene terephthalate (PET) bottles, maintained at the in situ temperature of the lake in a temperature-controlled room and incubated overnight under two pCO 2 and two nutrient concentrations (incubation overnight- Fig. 5). The + CO 2 level was maintained by constant bubbling throughout the night (12 h) from a gas tank at 750 ppm (Air Liquide, S.A.) to reach the pCO 2 predicted under the RCP 4.5 scenario 30 whereas the − CO 2 treatment was simulated by constant air bubbling to the samples (same as above) using an air pump. The pH of the samples was measured before and after 12 h of bubbling using a potentiometric titrator (Titrando 905, Metrohm, USA, Inc.) equipped with the Tiamo titration software v 2.0. The total CO 2 in the water samples was calculated from alkalinity and pH measurements 48 .
After the incubation period, subsamples coming from each 2-L-bottle and experimental treatment were subsequently placed into 50-mL quartz vessels, transported and incubated in the lakes for 4 h centred at local noon under two light qualities and two Mix treatments (exposure- Fig. 5). (2) overnight incubation, in which phytoplankton communities sampled were exposed overnight to both ambient and a nutrient pulse, under ambient and increased pCO 2 ; and (3) exposure, in which phytoplankton communities incubated under the nutrients and pCO 2 treatments mentioned were exposed in situ for 4 h centred on local noon to two solar radiation qualities: + UVR (> 280 nm) and − UVR (> 400 nm), and two light regimes: static (0.5 m depth) vs. fluctuating (moving up/down between 0 and 3 m depth).

Scientific Reports
| (2020) 10:19812 | https://doi.org/10.1038/s41598-020-76237-5 www.nature.com/scientificreports/ The Mix treatments were applied by using a customized mixing simulator equipped with a frequency-controlled DC motor (Maxon motor, Switzerland) that maintains a constant velocity (1 m every 4 min, ten cycles in total) throughout the incubations. Overall, all samples incubated at 0.5 m depth received mean irradiance values comparable to those of samples moving in the upper 3 m of the water column during the exposure period (UVR t test = 1.68, p = 0.17 ; PAR t test = 1.96, p = 0.12; mean values Table S1), although lake-specific differences existed. Thus, and due to the differential attenuation of solar radiation in the lakes tested, it was not possible to completely match both UVR and PAR in all lakes (Table S1). Both trays were placed ~ 2 m from the side of the boat, using an aluminium pole so that they were not shaded by the boat and had no interference from the shoreline during the incubation period.
A 2 × 2 × 2 × 2 full factorial design (in triplicate) was implemented for each lake with the following factors: (1) the UVR factor, with two qualities of solar radiation: − UVR (samples receiving only PAR, > 400 nm) with the quartz vessels covered with UVR-filter foil (UV-Process Supply Inc., IL, USA) and, + UVR (samples receiving UVR + PAR, > 280 nm) with uncovered quartz vessels. This treatment is intended for the evaluation of the net UVR effect because the experimental lakes are exposed to extreme UVR levels during spring-summer days (noon irradiances: ~ 6/30/70 µW cm −2 for 305/320/380 nm, respectively). Chemical variables. Samples for TP, TN and Si were placed in 300 mL PET bottles, frozen at − 20 °C, and analysed following standard protocols 48 . For DOC determinations, aliquots of 150 mL were filtered through precombusted Whatman GF/F filters (25 mm in diameter), placed in glass vessels, acidified with 100 µL of 1 N HCl (2% final concentration) and measured using a TOC analyser (Shimadzu, model 5000, Japan) 51 .

Biological variables.
For Chl a, 300-mL samples were filtered onto Whatman GF/F filters (25 mm in diameter), and stored at − 20 °C until analysis (Supplementary text S1). Phytoplankton abundance was determined following the Utermöhl method 52 from samples fixed with alkaline Lugol's (~ 1% vol/vol) preserved in 125-mL brown glass bottles (Supplementary text S2).
For primary production, 50-mL samples were inoculated with labelled NaHCO 3 (5 µCi; Perkin Elmer, Inc. USA) to measure inorganic 14 C incorporation 53 , and incubated in situ during 4 h centred at local noon (Supplementary text S3).
Data and statistical analyses. One-way analysis of the variance (ANOVA) was used to test significant differences among ln response ratios (lnRR; see Supplementary text S4) of the lakes tested. To test the effects of the TN:TP ratio, in situ temperature, and mean total irradiance received by communities on the P c response to UVR, CO 2 , Mix, Nut, and their interaction, a five-way analysis of the covariance (ANCOVA) was performed with UVR, CO 2 , Mix, Nut and lake, as fixed factors, and TN:TP ratio, in situ temperature, and mean total irradiance, as co-variables. We considered our lakes to be a fixed factor because all communities were exposed to the same experimental manipulation. We included the above mentioned co-variables in the ANCOVA analyses because they are environmental drivers that often operate on similar time scales as P c in surface waters and therefore could potentially modulate the P c responses to UVR, CO 2 , Mix, and Nut and their interaction. Prior to the ANOVA and ANCOVA analysis, assumptions of normality (by Q-Q plot residual analysis and Shapiro-Wilk's test) and homoscedasticity (by Levene's Equal Variance test) were checked. Homogeneity of regression slopes between Kd PAR and co-variables were checked through Pearson's correlation analysis, and linearity between P c and co-variables through dispersion plots. Differences among and within treatments and/or lakes were detected www.nature.com/scientificreports/ using a post hoc least-significant difference test. Finally, Student's t test was used to compare global mean irradiances received by samples during exposure to static and mixing treatments. Single and interactive UVR, CO 2 , Mix and Nut effects on P c were quantified using natural logarithm response ratios (lnRR) according to the corrected formulation of Harvey et al. 54 (Supplementary text S4). The relationship between the lnRR single, double, triple, and interactive drivers tested and the Kd PAR gradient on P c were assessed by polynomial regression analyses. We used non-linear regression fits because: (1) they explained a higher proportion of the total variance of the P c by Kd PAR than when using linear regression models (R 2 < 0.40 in all interactive effects); and (2) we obtained lower values of Akaike's information criterion (AIC) resulted in all interactions when compared with linear regression models (AIC polynomial [ranging between − 5.71 and − 25.71]; AIC linear [ranging between − 3.29 and − 17.26]). After regression analyses, assumption of normal distribution was checked through residual analyses.

Data availability
All data used in this study are included in the manuscript and supplementary information, and will be available under request to the corresponding author. www.nature.com/scientificreports/