The effect of increasing temperature and pCO2 on experimental pelagic freshwater communities

As the global climate is changing, average water temperatures and the supply of CO2 to water bodies are increasing. To determine how the effects of these changes on freshwater communities interact, we ran a month‐long factorial mesocosm experiment, in which we manipulated water temperature (heated, ambient) and pCO2 (preindustrial, ambient, future). We found that the total phytoplankton biomass responded positively to the pCO2 and temperature treatments but no interactive effects were detected. Green algae were positively affected by temperature and, over the course of the experiment, responded to pCO2 first positively, then negatively. Heterokonts, on the other hand, were unaffected by temperature but responded positively to pCO2. pCO2 enrichment also led to increases in seston C : N stoichiometry, although the experiment ended before we could observe any effects of pCO2 on the zooplankton community composition. Warming caused shifts in zooplankton community composition, primarily through higher abundances of the cladoceran Bosmina longirostris and the rotifer Conochilus unicornis, and lower abundances of the rotifer Polyarthra vulgaris. We found that, in contrast to the effects of temperature, which can be explained by temperature‐dependent plankton growth curves, the responses of algal groups to pCO2 enrichment were difficult to anticipate, despite the availability of priors from previous pCO2 enrichment experiments in the same lake mesocosms. We conclude that climate change‐induced increases in aquatic pCO2 and temperatures are likely to affect pelagic ecosystems, though further research in freshwater systems is needed before generalized claims regarding the simple and interactive effects of pCO2 can be made.

Around the world, climate change is having major impacts on precipitation patterns and lake water temperatures, along with related properties such as winter ice cover, evaporation, and mixing regimes (Woolway et al. 2020). However, the impacts of climate change on lake ecosystems are yet to be fully understood. Climate change is the result of increasing levels of greenhouse gases (primarily CO 2 , but also methane, nitrous oxide, and others) trapping heat in the atmosphere (IPCC 2013). While such rising temperatures can influence a range of aquatic biological processes in lakes, the direct effects of elevated CO 2 may also be important, yet are generally less well understood.
More fundamentally, while the rise of lake water temperatures worldwide is well documented (O'Reilly et al. 2015), less information is available regarding long-term trends of other lake properties shaping ecological processes. Although the few multidecade-long time series of aquatic partial pressures of CO 2 (pCO 2 ) in lakes feature more examples of increasing than decreasing pCO 2 over time, many more studies would be needed to make any generalizations (Couturier et al. 2022;Minor and Brinkley 2022). Unlike in marine systems, where aquatic pCO 2 is typically near equilibrium with atmospheric pCO 2 , freshwater systems are frequently in a state of disequilibrium (Cole et al. 1994). Although atmospheric pCO 2 also plays a role in controlling aquatic pCO 2 in lakes, the disequilibrium is caused by an imbalance between the rates of respiration and primary production, causing either the depletion or accumulation of aquatic pCO 2 (Cole and Prairie 2009), along with catchment characteristics which control the input of terrestrial carbon. In a global-scale analysis of 4902 lakes, it was found that the primary factor controlling lake pCO 2 is the concentration of dissolved organic carbon, that is, the substrate for bacterial respiration (Sobek et al. 2005). It is apparent that, at least in the Northern Hemisphere, dissolved organic carbon concentrations in lakes are increasing, resulting in higher levels of aquatic pCO 2 (Lapierre et al. 2013). This increase in dissolved organic carbon, also called "brownification", appears to be driven by a number of factors, including recovery from past acid deposition, land cover and use, and climate change and weather patterns (Kritzberg et al. 2019). Therefore, it is reasonable to assume that temperature and pCO 2 will simultaneously rise in many lakes. Many studies have investigated the effects of temperature and CO 2 on aquatic populations, communities, and ecosystems, yet few have focused explicitly on the potential interaction of these two factors.
Increasing temperatures are known to cause changes in phytoplankton and zooplankton community composition (Dupuis and Hann 2009;Huisman et al. 2018). In the most direct sense, maximum growth rates of populations are closely related to temperature for zooplankton (Gillooly 2000) and phytoplankton (Paerl et al. 2011). Among the phytoplankton, cyanobacteria generally have the highest temperature for optimal growth, followed closely by green algae, then by dinoflagellates, and finally diatoms (a class of heterokonts) (Paerl and Otten 2013). Shifts in the phytoplankton community composition can also affect the size-structures and composition of the zooplankton community; for example, increases in filamentous cyanobacteria can cause problems for the filtration apparatus of larger Daphnia species, whereas smaller species are less strongly affected (DeMott et al. 2001;Huisman et al. 2018). Conversely, shifts in the zooplankton community composition can also affect phytoplankton community composition because different zooplankton taxa can have different feeding preferences, most importantly, with respect to prey size (Elser and Carpenter 1988;Lewandowska et al. 2014). Furthermore, increasing amounts of total phytoplankton biomass are likely to promote zooplankton biomass due to increased food availability, whereas increasing amounts of zooplankton biomass are likely to cause a decrease in the phytoplankton biomass due to increased grazing.
Increasing aquatic pCO 2 can result in increased growth rates, biomass, and productivity of phytoplankton populations (Vogt et al. 2017;Hamdan et al. 2018), though which specific groups benefit most strongly from this increase in CO 2 remains a topic of debate (Huisman et al. 2018). Research on the algal physiology of carbon uptake is a major source of understanding on this topic. Briefly, RuBisCO enzyme, responsible for carbon fixation, that is, the rate-limiting step of photosynthesis, was found to be, by itself, highly inefficient (Morell et al. 1992). To compensate for this inefficiency, it was discovered that most phytoplankton groups have "carbon concentration mechanisms" (CCMs), a diverse array of adaptations that allow phytoplankton to actively take up carbon from surrounding water and saturate the RuBisCO enzyme (Reinfelder 2011). Some species have CCMs that allow them to take up bicarbonate (HCO 3 À ), which can be advantageous when aquatic CO 2 becomes depleted, the pH shifts to higher values, and even more dissolved CO 2 is converted to HCO 3 À (Nimer et al. 1997). In their literature review, Low-Décarie et al. (2014) found that cyanobacteria tend to have less efficient RuBisCO enzymes, but much more effective CCMs relative to other phytoplankton groups such as green algae and diatoms, leading to the hypothesis that green algae may benefit from increased aquatic pCO 2 concentrations. Indeed, this hypothesis has been supported in whole-lake experiments (Shapiro 1997) as well as from mesocosm and in vitro competition experiments showing that green algae tend to outcompete cyanobacteria under elevated CO 2 concentrations (Low-Décarie et al. 2011, 2015. Others, however, found that across 69 boreal lakes, aquatic pCO 2 does not influence community composition (Vogt et al. 2017), that in mesotrophic mesocosms, there are seasonally varying effects of CO 2 on the competition between heterokonts, green algae, and cyanobacteria (Katkov et al. 2020) or alternatively, that in mathematical models of eutrophic lakes, cyanobacterial blooms benefit from increased aquatic pCO 2 concentrations (Verspagen et al. 2014b;Ji et al. 2020). Furthermore, a series of laboratory experiments with seven desmid species (an order of green algae) demonstrated that many species can be expected to effectively compete for HCO 3 À with cyanobacteria (Spijkerman et al. 2005).
From a food web perspective, it is also clear that increased CO 2 concentrations can cause phytoplankton to have higher proportions of carbon relative to nitrogen and phosphorus and have effects on algal fatty acid composition, which collectively translate into reduced food quality for grazers (Urabe et al. 2003;Rossoll et al. 2012). Furthermore, increasing aquatic pCO 2 can cause the pH to decrease, which may slow growth rates of some zooplankton species, though this effect is often weaker than indirect, food quality effects (Urabe et al. 2003;Meunier et al. 2016). Freshwater zooplankton in particular have likely evolved to experience a wider range of pH concentrations than their marine counterparts due to the relatively high variability of pCO 2 within and across freshwater systems (Cole et al. 1994;Minor and Brinkley 2022). For example, diel cycles of net autotrophy during the day and net heterotrophy at night can readily result in a change of one or more units of pH (Maberly 1996).
The interaction between temperature and CO 2 , on the other hand, is much less studied. The best-understood aspect of this interaction is the way in which temperature modulates the effects of a CO 2 -induced reduction in food quality on zooplankton developmental rates. Three studies that have addressed this issue have found that temperature is an important modulator, but the specific relationship seemed to vary according to species (Persson et al. 2010;Malzahn et al. 2016;Garzke et al. 2020). For example, laboratory experiments with Daphnia magna found that decreasing food quality had a stronger impact on growth rates at higher temperatures (Persson et al. 2010). In contrast, laboratory experiments with the calanoid copepod Acartia tonsa showed that decreasing food quality had stronger impacts at low temperatures (Malzahn et al. 2016). Finally, a mesocosm experiment, also focused on A. tonsa, found that, in warm temperatures, increased aquatic pCO 2 promoted faster growth and developmental rates but greater mortality, while in colder temperatures, higher aquatic pCO 2 did not affect developmental rates but resulted in decreasing mortality rates (Garzke et al. 2020). In summary, studies find species-specific interactions between temperature and food quality, suggesting that warming and increasing pCO 2 can interactively affect zooplankton biomass and community composition in natural systems.
To investigate the combined effects of increasing CO 2 and temperature on the phytoplankton and zooplankton communities, we conducted a mid-fall mesocosm experiment in a mesotrophic lake. We chose the time of year because previous experiments performed at the same study site have shown significant effects of CO 2 on community composition during this period (Low-Décarie et al. 2015;Katkov et al. 2020). We tested the broad hypothesis that temperature and pCO 2 co-regulate planktonic communities through interactive effects on individual growth and food quality. Specifically, we make five predictions: (1) chlorophyll a (Chl a) biomass (as a proxy for total phytoplankton biovolume) will increase as a result of both temperature and CO 2 increases, possibly in a synergistic manner.
(2) Increasing aquatic pCO 2 will benefit green algae (Low-Décarie et al. 2015;Katkov et al. 2020). (3) Higher temperatures will benefit cyanobacteria and green algae, possibly at the expense of heterokonts, as expected from the optimal temperature for maximal growth of each group (Paerl and Otten 2013). (4) Food quality will decrease in response to increasing aquatic pCO 2 (measured as an increase in the seston carbon to nitrogen ratio) as expected from a number of laboratory experiments (Urabe et al. 2003;Meunier et al. 2016). (5) The zooplankton community composition will be affected by temperature and CO 2 via direct species-specific effects of temperature on growth rates (Gliwicz and Lampert 1990;DeMott et al. 2001;Huisman et al. 2018), via indirect positive effects of temperature and CO 2 mediated by increased phytoplankton abundance, and indirect, and via potentially interactive and species-specific negative effects caused by decreasing food quality (Persson et al. 2010;Malzahn et al. 2016;Garzke et al. 2020).

Experimental design
The experiment consisted of 18 mesocosms, crossing two levels of temperature (ambient and heated) with three levels of CO 2 (low, medium, and high), with three replicates. We aimed to generate a temperature difference of 2-3 C, and aquatic pCO 2 levels of 250, 400, and 1000 ppm. The aquatic pCO 2 levels fall within the range of values measured within the study system (see Study site and data collection), and were selected to represent states of equilibrium with the atmosphere at preindustrial, current and future conditions. Mesocosms assigned to the warming treatment were heated using a 300 W aquarium heater (Eheim) and those assigned to the ambient level were not. Heaters were placed near to bottom of the mesocosms to generate convection and ensure even heating of the water column. When submerged, the heaters were warm to the touch, but unlikely to produce enough heat to cause any significant levels of mortality due to contact with organisms. Other mesocosm (Hébert et al. 2023) and laboratory (Basu and Mackey 2022) experiments have used similar heaters and have not reported increased mortality in heated treatments. All mesocosms were bubbled with atmospheric air with altered levels of atmospheric pCO 2 for 2 min h À1 during the day, and for 2 min every 2 h between 22:00 h and 04:00 h. Bubbling was achieved by pumping air through weighted perforated tubing placed inside each mesocosm and connected by weighted "Tornado" tubing (CanadianPond.ca; Low-Décarie et al. 2015) to an air distribution network constructed with 1/2 in. polyethylene tubing inside the mesocosm platform. The low-CO 2 air was filtered through 2 L of soda lime (Fauna Marin Skim Breeze) to remove a fraction of CO 2 gas, medium-CO 2 air was untreated, and high-CO 2 air was enriched to 4500 ppm using pressurized CO 2 . The bubbling treatment had the added benefit of mixing the mesocosms, thus preventing stratification, and ensuring even distribution of the warming treatment and organisms throughout the water columns.

Study site and data collection
The study site was a floating platform located on Lac Hertel, Mont-Saint-Hilaire, Quebec, Canada. The platform was located 30 m offshore, near the deepest part of the lake. Lac Hertel is part of a UNESCO world heritage site, which includes the Gault Nature Reserve, managed by McGill University; the watershed of the lake is primarily comprised of a hilly old-growth forest. The lake itself is dimictic, with a maximum depth of 8 m, a mean depth of 4.7 m and a surface area of 0.31 km 2 (Goswami 1971;Rooney and Kalff 2003). At the start of the experiment, we found that the average total phosphorus concentration in the 18 mesocosms was 17.6 AE 0.2 μg L À1 (AE SEM) and the total nitrogen concentration was 0.321 AE 0.005 mg L À1 , which was comparable to the lake (Supporting Information Fig. S1a,b). However, as the experiment progressed, the nutrient concentrations in the mesocosms decreased to 10.2 AE 0.3 μg L À1 of total phosphorus and 0.293 AE 0.005 mg L À1 of total nitrogen by day 31. In contrast, nutrient concentrations in the lake increased over the same period of time, likely as a result of a fall overturn and mixing of nutrient-rich hypolimnetic water (Supporting Information Fig. S1a,b). The aquatic pCO 2 of the lake, measured 11 times over the course of this fall experiment around 10 am averaged out to 820 AE 70 ppm (AE SEM). In contrast, summer aquatic pCO 2 in the lake was found to decrease to levels below 50 ppm (Katkov et al. 2020). Previous studies have qualified the lake as soft water, bordering on hard water, with summer alkalinity ranging from 0.48 to 0.66 meq L À1 (Kalff 1972;Hem 1985). For additional information about the lake over the course of the experiment, see Supporting Information Fig. S1.
The mesocosms were constructed from 750 L, white polyethylene cylindrical tanks with a diameter of 0.79 m and a height of 1.4 m (Norwesco Inc.) by sawing off the tops and adding 25 by 30 cm shelving brackets (Everbilt) to the tops of the tanks to prevent the mesocosms from escaping through the 1 m wide rings installed on the floating dock. When installed and filled, the water depth in the mesocosms was 1 m and the volume was 624 L. The solid construction of the tanks (in lieu of polyethylene bags) was helpful to avoid the destruction of mesocosms by aquatic wildlife, to prevent leakage, and to reduce plastic pollution of the lake. Because our experimental design involved periodic bubbling of the mesocosms, stratification was highly unlikely. The mesocosms were filled with unfiltered lake water from a depth of 1 m using a electric centrifugal pump. The recently filled mesocosms were first sampled on 5 and 6 October 2020 (days À2 and À1 of the experiment), the temperature treatment was first applied on 7 October 2020 (day 0) and the CO 2 treatments started on 8 October 2020 (day 1). The experiment ended on 17 November 2020 (day 41).
Twice per week, between 09:00 h and 11:00 h, water temperature, pH, specific conductance and dissolved oxygen data were collected (in the mesocosms and in the lake close to the mesocosm platform) at a depth of 0.5 m using a YSI probe. Simultaneously, samples for estimation of aquatic pCO 2 and phytoplankton communities were collected. Although neither temperature, nor CO 2 treatments were interrupted for taking measurements or samples, mesocosms were not measured or sampled during the period of active bubbling.
Samples for aquatic pCO 2 were collected from a depth of $ 0.3 m using 60 mL syringes with a valve, which was closed immediately after collection to avoid any gas exchange with the atmosphere. Aquatic pCO 2 was measured using the headspace method (Cole and Prairie 2009). In summary, inside the sealed syringe, 30 mL of lake water were mixed with 30 mL of atmospheric air stripped of CO 2 with a soda lime column (Molecular Products). After equilibration, the air sample was injected into an infrared gas analyzer (PP Systems). Aquatic pCO 2 was then calculated from the pCO 2 of the air sample measured by the gas analyzer, the temperature of the mesocosm and the final sample temperature by way of Henry's law (Cole and Prairie 2009).
Water samples for phytoplankton analysis were collected using 30 mL polyethylene tubes by introducing them into the water up-side down and inverting to fill them at a depth of $ 0.3 m. The tubes were covered in electric tape to avoid exposure to sunlight and stored in a cool environment for a maximum of 4 h to avoid any significant levels of phytoplankton mortality. The samples were analyzed using a bench-top Fluoroprobe (bbe Moldaenke, GmbH) using default parameter settings to characterize the Chl a biomass and the chemotaxonomic community composition (green algae, heterokonts, and cyanobacteria) in a dark room. Fluorometers such as the Fluoroprobe use specific wavelengths to excite molecules of Chl a and accessory pigments and then measure the fluorescence emissions of these molecules, which have known and specific wavelengths (Catherine et al. 2012).
Three times over the course of the experiment, zooplankton and nutrient samples were collected. Water for the zooplankton samples was collected using a 2.2 L Ruttner sampler from the center of each mesocosm and from the lake, right off the mesocosm platform (at a depth of 0.5 m). We concentrated the zooplankton in ethanol by filtering 11 L of mesocosm water through a 30 μm mesh sieve. The filtered water was placed back into the mesocosm. The entire sample was used to identify and count crustacean individuals. Due to very high abundances of rotifers, individuals from this group were counted and identified in a subsample corresponding to 10% of the total sample volume. In either case, a Nikon SMZ800 dissecting microscope, and a Nikon Eclipse TE2000-S inverted microscope were used (Thorp and Covich 2001;Hudson and Lesko 2003;Haney et al. 2013) (species list: Supporting Information Table S1). For all adult crustaceans and the five most abundant rotifer species, abundance data were transformed to biomass using species-or genus-specific estimates of individual dry mass (Pauli 1989;Anderson and Benke 1994;Hébert et al. 2016).
For total and dissolved phosphorus and nitrogen in each mesocosm and the lake, $ 1 L samples were collected from a depth of 0.3 m. Subsequently, duplicate volumes of 40 mL were subsampled into acid-washed glass tubes. Total fractions did not require any filtration, while dissolved fractions were filtered through a GF/F filter using a manual syringe with a reusable filter holder attachment. The filters were preserved for analysis of the particulate stoichiometry. Following digestion with potassium persulfate and the addition of an ammonium molybdate solution, total phosphorus concentrations were measured using colorimetric detection with a spectrophotometer at 890 nm (Wetzel and Likens 2000). Total nitrogen concentrations were measured using a continuous flow analyzer (ALPKEM Flow Solution IV; OI Analytical) using an alkaline persulfate digestion method, coupled with a cadmium reactor (Patton and Kryskalla 2003). The relative carbon and nitrogen masses of each GF/F filter were determined using a Carlo Erba 2500 elemental analyzer and were later used to calculate the C : N ratios of the seston.

Statistical analyses
For measurements taken twice per week (temperature, aquatic pCO 2 , Chl a, phytoplankton community composition), a linear mixed model was fitted to simple and interactive effects of CO 2 treatment, temperature treatment and time with each individual mesocosm as a random factor using the R package lme4 1.1.30 (Bates et al. 2015). Note that we considered exposure time to be a factor, which allowed us to compare the treatment effects at different levels for each sampling date and to avoid making any assumptions about the relationship between response variables and time. Using the fitted model, ANOVA statistics with p values were calculated using the R package lmerTest 3.1.3 (Kuznetsova et al. 2017). For factors, or interactions between factors that were found to be significant (p < 0.05), differences between the estimated marginal means of the model were used to determine which treatments were significantly different from each other (p < 0.05) using the R package emmeans 1.8.1.1 (Lenth 2021). If the interactions included the exposure time, the comparisons were made separately for each day, and we reported the days on which effects were significant.
For univariate measurements taken three times over the course of the experiment (seston stoichiometry and zooplankton biomass), for each date, a linear model was fitted, with simple and interactive effects of temperature and CO 2 treatments as predictors. As for the mixed models, after calculating the ANOVA statistics for each model fit, significant factors (p < 0.05) were analyzed further by comparing estimated marginal means.
To detect shifts in the zooplankton community composition, we ran a redundancy analysis (RDA) on Hellingertransformed community data with temperature and CO 2 treatments as constraining variables for each of the three sampling dates using the R package vegan 2.6.2 (Oksanen et al. 2020). For each RDA, we calculated an ANOVA table to determine which factors had a significant effect on the zooplankton community composition. All analyses were performed in R version 4.1.2 (2021-2111-01).

Treatment effectiveness
Overall, we found that the CO 2 and temperature treatments were effective and independent of each other. On average, across all days posttreatment application (n = 10), high-CO 2 treatments had mean (AE SEM) aquatic pCO 2 of 2000 AE 100 ppm; medium-CO 2 treatments had 680 AE 20 ppm; and the low-CO 2 treatments had 430 AE 10 ppm (Fig. 1a). For comparison, global average atmospheric concentrations at the time of the experiment were approximately 410 ppm (Lan et al. n.d.). The absolute aquatic pCO 2 values attained in each treatment level were slightly higher than our target values but the differences among treatments were as expected and significant for nearly every sample date of the experiment. The Fig. 1. The effect of treatments on (a) aquatic pCO 2 and (b) temperature (y-axis) over time, expressed as the day since the start of the experiment (x-axis) in the mesocosms. In the "CO 2 Summary" panel, all lines are colored by CO 2 treatment (see legend), thick lines represent averages of different CO 2 treatments regardless of temperature treatment, thin lines represent averages for each CO 2 and temperature combination, the "*" symbol represents days where at least one significant difference (p ≤ 0.05) between different CO 2 treatments were found (see text for details), and the "-" symbol represents days where no significant differences ( p > 0.05) between different CO 2 treatments were found. Note that no sampling events took place between days 31 and 41. The interpretation for the "Temperature Summary" panel is analogous to the "CO 2 Summary", but for temperature instead of CO 2 treatments. Overall, we find that both treatments had the desired effects. For a comparison with the lake, see Supporting Information Fig. S1e,f. temperature in heated mesocosms was, on average, 2.7AE0.4 C higher than in unheated mesocosms (Fig. 1b). Over the course of the experiment, temperatures decreased, but remained significantly different between temperature treatments across nearly all days. Contrary to our expectations, pH was not affected by the CO 2 treatment (Supporting Information Fig. S2). This could be attributed to low specific conductance of the lake water, which averaged to 78.2 AE 0.1 μS cm À1 (AE SEM) in the mesocosms, as this makes pH measurement less accurate (Busenberg and Plummer 1987) but was not accounted for during sampling. On day 15, a single heater became unplugged and caused average temperatures to drop in one of the medium-CO 2 mesocosms (Fig. 1b). The heater was immediately plugged back in, and the treatment was rapidly restored.

Phytoplankton community
We found that differences in total Chl a (measured as a proxy for phytoplankton biomass) were controlled by the independent effects of CO 2 and temperature, both of which varied by date ( Fig. 2a; Table 1). We found that high pCO 2 mesocosms had significantly higher Chl a concentrations than low pCO 2 mesocosms on days 19, 24, and 27 of the experiment. Temperature caused Chl a to significantly  Table 1. Fixed effects of the ANOVA with respect to the response of total chlorophyll a. Interactions between terms are denoted with ":", "Sum Sq" are type III sums of squares, "Mean Sq" are the mean squares, "NumDF" and "DenDF" are the degrees of freedom of the numerator and denominator terms of the F-test, "F-test" is the result of the F-test, and "Pr(> F)" denotes the p value and "Signif" denotes the significance of the p-value (*p < 0.05; **p < 0.01; ***p < 0.001).

Term
Sum  We also found that the increases in Chl a biomass in response to increasing aquatic pCO 2 were reflected in decreasing concentrations of dissolved nitrogen and phosphorus, and, to a lesser extent, total phosphorus, but not of total nitrogen (Supporting Information Fig. S3).  The biomass of green algae was affected by heating and, independently, CO 2 concentrations, though both effects varied according to the date ( Fig. 2b; Table 2). On day 5, heating had a negative effect on green algal biomass, whereas on days 24, 27, 30, and 41, the effect was positive. The effect of CO 2 was more complex: on day 19, we found a significantly higher green algae biomass in high-and medium-CO 2 mesocosms, compared to low-CO 2 mesocosms. On the other hand, on day 41, we found significantly lower green algae biomass in the high-and medium-CO 2 mesocosms, compared to the low-CO 2 mesocosms. See Supporting Information Tables S3.1, S3.2 for contrasts.
Biomass of heterokonts was found to be principally controlled by the CO 2 treatment, though the effect varied by day ( Fig. 2c; Table 3). On days 19, 24, 27, 30, and 41, we found significantly higher biomass of heterokonts in the highvs. the low-CO 2 mesocosms. See Supporting Information Tables S4.1, S4.2 for contrasts.
Cyanobacteria, which comprised a very small proportion of the phytoplankton community by mid and late fall, were also affected by temperature on certain days ( Fig. 2d; Table 4). Cyanobacteria biomass was significantly higher in heated mesocosms on days 5 and 12. See Supporting Information Tables S5.1, S5.2 for contrasts.

Seston stoichiometry
We found some support for our prediction that the seston stoichiometry was affected by CO 2 supply. Prior to the application of treatments, and mid-way through the experiment (day 19), the ratio of particulate C : N in mesocosms was not significantly affected by either treatment or their interaction. On the third (and last) sampling (day 31), on the other hand, we found that CO 2 had a significant effect on the seston C : N ratio (ANOVA: F 2,10 = 17.6, p < 0.001), with an increase of 15% in the high-CO 2 relative to low-CO 2 mesocosms (p = 0.0019; Fig. 3).

Zooplankton community
Throughout the experiment, we found 13 distinct rotifer species, one of which was present in both colonial and individual forms, 13 distinct crustacean species, copepodites, and nauplii, for a total of 29 zooplankton taxa. Of these groups, only three rotifer species (Keratella cochlearis, Polyarthra vulgaris, and Ploesoma truncatum), one cladoceran species (Bosmina longirostris), copepodites, and nauplii were present in all mesocosms on all three sampling dates. We found that temperature had a significant effect on the zooplankton community composition on day 15 of the experiment (p < 0.05), though this effect mostly dissipated by the next sampling date, on day 31 of the experiment (Fig. 4). B. longirostris and Conochilus unicornis were positively associated with the heated temperature treatment, whereas P. vulgaris was associated with colder temperatures. None of the 27 other categories showed any statistically significant responses to the temperature treatment. Furthermore, CO 2 treatments had no statistically significant effects on the zooplankton community composition. We also did not find any significant effect of temperature or CO 2 on the total zooplankton biomass (Fig. 5), nor did we find a significant correlation between zooplankton and phytoplankton biomass. However, we note that in the lake, zooplankton biomass greatly increased between day 15 and 31, compared to modest increases, or decreases in the mesocosms (Fig. 5). For a comparison of the zooplankton community composition in the lake compared to the mesocosms, see Supporting Information Text S1, where we show that the filling of the mesocosms prior to the start of the experiment caused the community composition to diverge from that of the lake, that by day 15, the lake community composition was still significantly different from the mesocosms, yet more similar to that of the ambient than the heated mesocosms, and that by day 31, the mesocosm and lake community compositions were no longer significantly different.

Discussion
We found that aquatic pCO 2 and temperature increases had independent effects on the phytoplankton community, although the effects were not consistent across all the days of the experiment. On days when aquatic pCO 2 influenced phytoplankton community biomass, the effect was positive. This result is slightly surprising because previous experiments in Lac Hertel concluded that increased nutrient addition was essential for aquatic pCO 2 to have a positive effect on phytoplankton biomass (Low-Décarie et al. 2015;Katkov et al. 2020). It is possible that, because we used a model that allowed us to test the effect for each sampling date, we were able to detect transient effects that could have been missed in previous studies. In addition, in contrast to Low-Décarie et al. (2015) and Katkov et al. (2020), this study featured a low aquatic pCO 2 treatment, intended to mimic preindustrial levels of atmospheric pCO 2 , which is involved in most of the significant CO 2 effects. Nevertheless, several studies have also found that primary production and Chl a biomass correlate with aquatic pCO 2 (Jansson et al. 2012;Shi et al. 2015;Vogt et al. 2017). The temperature effect, when present, also had a positive effect on phytoplankton community biomass as would be expected in a community dominated by green algae (as was the case in the mesocosms), which typically have growth optima near 20 C (Paerl et al. 2011).
In our experiment, we did not observe synergistic interaction effects between the CO 2 and temperature treatments. The evidence from previous, marine studies is ambiguous. A microcosm study investigating a North Atlantic spring bloom community did find that increased temperature and aquatic pCO 2 had a synergistic effect on Chl a biomass (Feng et al. 2009). However, a similar study found that aquatic pCO 2 and temperature had different effects at two different sites of the Bering Sea: a deep-water site and one in the middle of a continental shelf (Hare et al. 2007). Specifically, a synergistic effect on Chl a was observed at the deep-water site, whereas, at the shelf site, increased temperature, aquatic pCO 2 , and the combination of both all had comparable negative effects on Chl a (Hare et al. 2007). Adding to the puzzle is a third microcosm experiment in the South China Sea which showed that, although the Chl a response to increased temperature was positive at ambient aquatic pCO 2 , this was not the case at 1000 ppm aquatic pCO 2 (Gao et al. 2017). To our knowledge, no such studies have been conducted in freshwater systems. However, evidence from marine systems suggests that the interactive response of Chl a to changing temperature and aquatic pCO 2 appears to be highly system-specific, and likely depends on the community composition (Gao et al. 2012). In addition, we found that in one instance (sampling day 24), the positive effect of increasing aquatic pCO 2 and temperature co-occurred on the same day, meaning that independent, additive effects should be considered.
In terms of phytoplankton community composition, we were surprised to find that increasing aquatic pCO 2 did not have a consistently positive effect on the biomass of green algae as reported in previous fall experiments in Lac Hertel (Low-Décarie et al. 2015;Katkov et al. 2020). On only two sampling days, we observed a significant effect on green algae, one with a positive, and another with a negative effect of increasing aquatic pCO 2 . Instead of green algae, heterokonts benefited more strongly from increasing aquatic pCO 2 in the second half of the experiment (Fig. 2b-d and Supporting Information Fig. S4). This also contrasts with several marine studies which found shifts away from heterokonts in response to rising aquatic pCO 2 or acidification (Hare et al. 2007;Petrou et al. 2019). In our experiment, however, we did not find a significant impact of aquatic pCO 2 on pH (Supporting Information Fig. S2), though small effects may have been missed because the readings were taken in cold, low salinity water. Regardless, some studies have also found that community composition can, in response to high aquatic pCO 2 , shift toward heterokonts or remain unchanged (Kim et al. 2006;Katkov et al. 2020). Another study also reported that growth rates of different strains of the marine coccolithophore Emiliania huxleyi responded differently to changes in carbonate chemistry (Langer et al. 2009). Thus, it seems likely that species-and strain-specific responses to changing aquatic pCO 2 (and concomitant acidification) might be responsible for the inconsistencies that we and others encountered when it comes to predicting community composition shifts among broader algal taxonomic groups.
In terms of temperature, we found that cyanobacteria, which represented a small fraction of the community, initially benefited from the warming treatment, though seasonally decreasing temperatures over the course of the experiment likely prevented their proliferation. This finding supports the idea that climate change is likely to prolong and exacerbate cyanobacterial blooms (Huisman et al. 2018). Green algae also benefited from increased temperature in the second half of the experiment. Green algae tend to have higher growth rates at warmer temperatures, which can explain this response (Paerl et al. 2011). The initial heating, however, appeared to have a brief negative effect, possibly due to the shock caused by a rapid temperature change. Overall, while heating had no effect on heterokonts, the positive effects on green algae and cyanobacteria resulted in a net community biomass increase with warming.
Because enriched aquatic pCO 2 concentrations led to lower dissolved nitrogen concentrations, and to shifts in the phytoplankton community composition, it is difficult to determine the exact cause of the increase in C : N ratios observed on day 30 of the experiment. Most phytoplankton species tend to have flexible stoichiometry (Klausmeier et al. 2008), so that a shift in community composition affecting C : N stoichiometry seems to be the least likely explanation. Nevertheless, because most studies interested in flexible stoichiometry focus on N : P ratios, not C : N, this option cannot be definitively excluded. A laboratory and modeling study determined that pCO 2 has the greatest impact on phytoplankton stoichiometry when nutrients are limiting, and the greatest impact on primary production when nutrient availability is high (Verspagen et al. 2014a). Thus, it seems fitting that following a CO 2induced increase in biomass, leading to reduced nutrient availability, we would observe an increase in the C : N ratio. This can explain why the shift in N : P took between 16 and 30 d to become apparent, compared to incubation experiments that observed the same effect after 5 d or fewer (Burkhardt et al. 1999;Losh et al. 2012). If this delay was simply due to slow rates of resource uptake, we would expect a more gradual shift in C : N ratios, rather than the sudden regime shift that we observed. Regardless of the mechanism, the slow C : N response could explain why aquatic pCO 2 did not have a significant effect on the zooplankton community biomass or composition via food quality effects (Meunier et al. 2016). The observed shifts in C : N ratios are less marked, but still comparable to those reported in other studies (e.g., Schoo et al. 2012;Meunier et al. 2016), indicating the potential for a measurable effect of food quality on zooplankton.
Indeed, the zooplankton community composition was only affected by temperature, not aquatic pCO 2 . Interestingly, temperature had an effect on day 15 after the start of the experiment, but this effect seemed to have mostly dissipated by day 31. While populations of the cladoceran B. longirostris and the rotifer C. unicornis benefited from the heated treatment, the rotifer P. vulgaris had higher counts at ambient temperature. This finding is consistent with previous findings with regard to the effect of temperature on these two species. Though we are not aware of any studies focused specifically on the effects of temperature on B. longirostris, another cladoceran, Daphnia rosea was found to have an optimal filtering rate at 20 C (Burns and Rigler 1967). A study of Lake Gosmere of the English Lake District found that C. unicornis were most abundant at temperatures of 8.5-16.0 C (Elliott 1977). In oligotrophic Yellow Belly Lake, Idaho, USA, clearance rates of C. unicornis decreased 10-fold as temperatures dropped from 14 C to 7 C (Armengol et al. 2001). In contrast, the growth rate of P. vulgaris appears to peak at lower temperatures, between 10 C and 20 C, and possibly as low as 5 C (Buikema et al. 1978). In summary, it appears that the shifts in the zooplankton community can be attributed to species-specific optimal temperatures for growing or grazing.
A key limitation of this study is that we do not have a firm handle on possible bottom-up and top-down effects, that is, indirect food web effects of the treatments via food availability or grazing. Because total zooplankton biomass was not affected by any of the treatments, it is unlikely that top-down effects on phytoplankton biomass differed across treatments. On the other hand, a bottom-up response of zooplankton biomass to the CO 2 -stimulated increase in phytoplankton may have been negated by a reduction of algal food quality (increased C : N ratio). Top-down or bottom-up effects of aquatic pCO 2 on community composition are unlikely because the zooplankton community composition did not respond to changes in aquatic pCO 2 . However, although species-specific temperature optima provide a more parsimonious explanation, bottom-up and top-down effects of temperature on planktonic community compositions cannot be ruled out. Furthermore, we did not account for the effects of potential periphyton growth on the inside of the mesocosms in our analysis. Although periphyton was not present in any visually striking amount, it may have had an effect on the ecological dynamics within the mesocosms via depletion of available nutrient resources for phytoplankton and, indirectly, zooplankton, which could explain the observed decreases in nutrient concentrations and plankton biomass over time (Supporting Information Fig. S1a,b,e,k). In contrast, the fall overturn of the water column in the lake caused a replenishment of nutrients in the epilimnion, causing our experiment to diverge from the lake environment over time. Finally, we have to acknowledge that our experimental mesocosms are not perfect representations of the conditions and dynamics as they occur in the real lake. We found differences in nutrient dynamics and plankton composition even between the ambient treatments and the lake, the potential implications of which we discuss in the Supporting Information Material (Supporting Information Text S1). Furthermore, when it comes to temperature, we need to take into account that our experimental intervention (resulting in +2.7 C above ambient) occurred during a period of an overall loss of 8 C in water temperature. It is obvious that many of the findings in our experiment will exclusively apply to this period of rapid temperature decrease and cannot readily be transferred to periods of rising temperatures in the spring or comparatively static conditions during the summer months.
Despite some limitations, mainly concerning the mechanistic explanation for some of the observed effects, our study provides a much-needed first step in understanding how pelagic freshwater systems will, and probably already are, responding to aquatic pCO 2 enrichment combined with increasing temperatures. Although we did not find any evidence for synergistic interaction between the two factors, we cannot make any generalizations since this is, to our knowledge, the first such study in any natural freshwater system. The additive effects of temperature and aquatic pCO 2 that we found highlight the independent contributions of both factors in determining phytoplankton biomass and shaping community composition. The effect of aquatic pCO 2 on phytoplankton community composition was particularly puzzling since the results were opposite to what was found in the same lake, in comparable studies conducted at the same time of year just 5 and 8 yr prior. A tentative explanation is that responses to aquatic pCO 2 enrichment require higher taxonomic, and perhaps even intra-specific genotypic resolution. As nutrients became limiting in pCO 2 -rich mesocosms in response to increased phytoplankton biomass, we found increased seston C : N ratios, indicating a decrease in algal food quality, albeit, too late to detect a response in the total biomass or species abundances within the zooplankton community. Nevertheless, this finding indicates that such a potential exists in natural freshwater systems. Although the effects of temperature on both phytoplankton and zooplankton communities could generally be explained by temperature-optimum curves, we could not exclude the possibility of bottom-up and top-down effects. Consequently, we encourage further in situ experimental studies to explore this interaction in different seasons, systems, and on longer time frames to have a better chance of capturing the indirect effects pCO 2 and temperature on freshwater ecosystems.

Data availability statement
The data and source code supporting this article are openly available in a GitLab repository at: https://gitlab.com/ekatko/ CO2xTempMesocosm.