Evidence of partial thermal compensation in natural phytoplankton assemblages

Whether phytoplankton growth is solely constrained by temperature (hotter is better) or compensated by thermal adaptation is still under debate. We measured the temperature sensitivity of natural phytoplankton communities at both short‐term and seasonal timescales using temperature manipulation experiments. The activation energy across communities (mean ± SE: Ei = 0.51 ± 0.12 eV, Q10 = 1.98) is significantly lower than that within communities (Ea = 0.80 ± 0.10 eV, Q10 = 2.80). Moreover, using a larger dataset of phytoplankton growth rates measured in (sub)tropical waters, we estimated the across‐community activation energy as 0.33 ± 0.06 eV (Q10 = 1.56), which is also lower than Ea. Our study is the first to suggest the “hotter is partially better” for natural phytoplankton communities, indicating that the phytoplankton communities can show some thermal adaptation capability. Our results highlight the importance of incorporating the differential temperature sensitivities at different timescales into the biogeochemical models to better evaluate how marine ecosystems will respond to climate changes.

Marine phytoplankton contribute to about half of global primary production and are critical in the ocean biogeochemical cycles (Falkowski et al. 1998). Reliably quantifying their responses to the projected ocean warming is vital to predicting the effect of climate change on marine ecosystem functioning, such as CO 2 uptake and sequestration (Vasseur and McCann 2005;Fussmann et al. 2014). The effect of temperature on marine phytoplankton is characterized by thermal performance curves (TPCs), which depicts the physiological response of a single phytoplankton taxon to the temperature at short-term timescales (minutes to days). However, the temperature sensitivity of phytoplankton in terms of activation energy (E, eV) or Q 10 applied in most biogeochemical models is derived from the relationship of the maximum growth rates against temperature across various phytoplankton species, encompassing the difference in the adaptation to diverse habitats at longer timescales (months, seasons, or years) (Eppley 1972;Yool et al. 2013;Laufkötter et al. 2015).
Short-term thermal response of phytoplankton growth is typically unimodal with negative skewness, depicted by the TPC (Fig. 1a; Thomas et al. 2012). The growth rates of phytoplankton increase with rising temperature up to the maximum (μ max ) at the optimal temperature (T opt ) and rapidly decrease when temperature surpasses T opt (Schoolfield et al. 1981; Angilletta 2006). The thermal traits of phytoplankton growth (e.g., T opt and μ max ), which are embodied in the TPC and have the imprint of previous environments, are used to link the short-term thermal response to the long-term response involving adaptation (Knies et al. 2009;Kontopoulos et al. 2020). Nonetheless, the adaptation patterns revealed by the relationship between thermal traits (T opt and μ max ) remain debatable in the evolutionary biology literature. The "hotter is better" hypothesis posits that μ max should increase with T opt in a similar manner to the short-term response within one certain species as the organisms residing in cold environments cannot overcome the thermodynamic constraints of their biochemical kinetics and inevitably have a lower μ max than those from warm environments (Frazier et al. 2006;Knies et al. 2009). Thus, the short-term temperature sensitivity of phytoplankton growth rate within species should not differ from the long-term temperature sensitivity across species (see Fig. 1b). This hypothesis is implicit in estimating temperature sensitivity from the Eppley curve (Eppley 1972), which is usually applied in marine biogeochemical models (Yool et al. 2013;Laufkötter et al. 2015).
However, some studies argued that the adaptation of biochemical machinery could reduce or minimize the influence of thermodynamic constraints, allowing the organisms in cold environments to upregulate their performances (Clarke and Fraser 2004;Angilletta 2009). An extreme hypothesis is the "biochemical adaptation," which states that the maximum performance of cold-adapted organisms can be upregulated to the levels of the warm-adapted organisms-consequently, the μ max should be independent of temperature (see Fig. 1c). An intermediate scenario is also posited as the "partial compensation" hypothesis (i.e., "hotter is partially better") in which the influence of adaptation is weaker than what the "biochemical adaptation" hypothesis predicts, resulting in a weaker thermal response of μ max to the increase in T opt than the physiologically thermal response within species (see Fig. 1d). Recent studies have found that the adaptation of phytoplankton to the thermal environments is in the absence of hard "hotter is better" constraints but more aligned with the "partial compensation" hypothesis (Barton and Yvon-Durocher 2019;Kontopoulos et al. 2020). Thus, the short-term thermal response would be partially compensated by adaptation at the species levels, leading to a lower across-species temperature sensitivity than within-species (Barton and Yvon-Durocher 2019). Nevertheless, it remains unexplored whether this pattern also applies to natural phytoplankton assemblages as cultivated species or strains in the laboratory cannot fully represent the communities in the real ocean.
In this study, we aim to evaluate the thermal response of phytoplankton community at both short-term and long-term (seasonal) scales. We used two datasets on the growth rate of natural phytoplankton community estimated by the dilution technique, which is the only widely used method for measuring phytoplankton specific growth rates in situ (Landry and Hassett 1982;Laws 2013). We compiled the data of short-term temperature manipulated dilution experiments which estimated the growth rate of the same phytoplankton community at different temperatures. We also calculated the long-term activation energy from a dataset of phytoplankton growth rates in the (sub)tropical regions where the optimal temperature of phytoplankton communities can be approximated by in situ temperature (Thomas et al. 2012;Liu et al. 2019). By analyzing the thermal response of phytoplankton growth rate within vs. across communities, we test whether the temperature sensitivity of phytoplankton growth rate across communities should be lower than that within communities but still significantly positive as described in the "partial compensation" hypothesis.

Data compilation and description
We compiled the data of short-term temperature manipulated dilution experiments which measured the growth rate of natural phytoplankton communities at designated temperatures and estimated the short-term thermal response (within 1 d) of a phytoplankton community (Table S1). The experiments were conducted monthly for a year at three stations ( Fig. 2), which allow us to estimate the thermal responses across communities at a longer timescale (i.e., seasonal). The details of the experiments are in Chen and Liu (2015) and Liu et al. (2019). Briefly, in each experiment, five sets of dilution experiments were conducted, and the bottles were incubated at five temperatures (in situ temperature T, T À 5 C, T À 3 C, T + 3 C, T + 5 C). For each dilution experiment, particle-free seawater was prepared by filtering through a 0.2 μm filter capsule into 1.2-liter PC bottles and mixed with the natural seawater to a designated proportion (15% natural seawater). Another pair of 1.2-liter PC bottles were filled with natural seawater (100% natural seawater). Nutrients (NO 3 À : 10 μmol L À1 , PO 4 3À : 1 μmol L À1 in final concentration) were added into all bottles to ensure enough nutrients for phytoplankton growth. All bottles were incubated at the designated temperature for 24 h under simulated light conditions. At the beginning and the end of the incubation, the chlorophyll a concentration (Chl a) of each bottle were measured (Supplementary Information). By assuming an exponential growth model for phytoplankton communities, the net growth rate of phytoplankton (k) in each bottle was then calculated as k = (1/t) ln (P/dP 0 ), where P 0 and P are the initial and final Chl a in each bottle, d is the dilution factor, and t is the incubation time. The growth rate of phytoplankton community was estimated as the intercept of the linear regression of the net growth rate (k) and dilution factor (d).
As such short-term experiments were conducted in the subtropical waters, where the T opt of phytoplankton has been found to be close to the environmental temperature (Thomas between biological rates and temperature (the thermal performance curve). T opt is the temperature at which the rates are maximal (μ max ), E a is the temperature sensitivity describing the curve rise below T opt . (b) The "hotter is better" scenario for thermal adaptation. The curves represent the thermal responses of different communities. Under such a scenario, the μ max increases with T opt with the same temperature sensitivity (E a ) due to the thermodynamic constraint. (c) The "biochemical adaptation" hypothesis, whereby the long-term adaptation will override the effects of thermodynamic constraint such that μ max is invariant with T opt . (d) The "partial compensation" hypothesis, whereby the μ max increases with increasing T opt with a lower temperature sensitivity because of the partial thermal adaptation.  Table S2). In this dataset, phytoplankton growth rates were estimated by the dilution experiments described above and conducted in various locations and seasons (Supplementary Information), allowing us to evaluate the thermal responses across various phytoplankton communities at seasonal timescales. In these experiments, nutrients were also added to the experimental bottles to ensure sufficient nutrients for phytoplankton growth. Considering the effect of light intensity, we selected the data from experiments conducted in surface layers, and the corresponding photosynthetically active radiation (PAR, mol photons m À2 d À1 ) was collected from the Goddard Earth Sciences Data and Information Service Center (http://disc.sci.gsfc.nasa.gov/).

Short-term temperature sensitivity within the community
Based on the dataset of short-term experiments, the TPC for the growth rate of a phytoplankton community was quantified using a modified Sharpe-Schoolfield equation (Sharpe and Demichele 1977;Schoolfield et al. 1981) following the previous study (Chen and Laws 2017;Padfield et al. 2017): where μ (d À1 ) is the growth rate of phytoplankton at temperature T (K), μ c is the growth rate normalized to a reference temperature T c (here T c = 15 C), k b is the Boltzmann's constant (8.62 Â 10 À5 eV K À1 ), T opt is the optimal temperature at which the growth rate is the maximal, E a is the activation energy (eV) within the same community which describes how fast the growth rate increase with increasing temperature below T opt , E h is deactivation energy induced by high-temperature inactivation. We fitted Eq. 1 to each phytoplankton community by the Levenberg-Marquardt nonlinear least-squares minimization algorithm, which was implemented using the nlsLM function in minpack.lm package in R (R Core Team 2017). The optimization process during the fitting was enforced by strict convergence criteria according to Kontopoulos et al. (2020). The fits were then filtered by criteria to ensure reliable parameter estimates. We removed the fits with E a < 0.01, E h > 10, or E h < 0.01 to ensure that the TPCs are unimodal. The μ max was calculated based on the estimated parameters by setting the temperature as T opt . As the estimates of thermal traits were affected by the different phytoplankton communities among months, we estimated the average value of each parameter in Eq. 1 using the nonlinear mixed-effect models allowing random effects on each parameter. The Saemix package based on the stochastic approximation expectation maximization algorithm was used for this analysis (Comets et al. 2017; Supplementary Information).

Long-term temperature sensitivity across communities
The temperature sensitivity across phytoplankton communities (E i ) was estimated based on the Boltzmann-Arrhenius equation (Chen et al. 2012;Padfield et al. 2017). As the phytoplankton growth may also be affected by size structure which changes with phytoplankton biomass (Chen et al. 2012), we also included the effect of biomass that represented by Chl a (Chl): where μ max is the maximal growth rate at optimal temperature T opt estimated from Eq. 1 based on the dataset of short-term experiments; E i is the activation energy across communities; other symbols are the same with Eq. 1. To corroborate the above estimate of E i , we used the larger dataset of in situ phytoplankton growth rates by assuming that the T opt of phytoplankton could be approximated by the environmental temperature and the growth rates were close to μ max in (sub)tropical waters (Thomas et al. 2012;Liu et al. 2019). We used generalized additive models (GAMs) to estimate E i 0 . We considered the effect of biomass (Chl) and light intensity (PAR) on phytoplankton growth rate by constructing the following model: where μ n is the phytoplankton growth rate estimated at in situ temperature (T in situ ), μ c 0 is the normalization constant, and s is the smooth function (thin plate regression splines) in GAMs. As the growth rates were estimated under nutrientenriched conditions, we did not add the effect of nutrients into the model. The mgcv package in R was used for the GAMs analysis (Wood 2006).

Comparing short-and long-term temperature sensitivities
To test whether the across-(E i or E i 0 ) and withincommunities (E a ) activation energies were significantly different, we used bootstrap hypothesis testing following Efron and Tibshirani (1993)

Temperature sensitivity (E a and E i ) estimated by temperature manipulated experiments
The phytoplankton community growth rates in most experiments showed unimodal patterns with increasing temperature and were fitted by the modified Sharpe-Schoolfield equation (Fig. 3a). The across-community activation energy E i (mean AE SE: 0.51 AE 0.12 eV, the same below, equivalent to Q 10 of 1.98; Fig. 3b) was significantly lower than the within-community E a estimated by the nonlinear mixed-effect model (0.80 AE 0.10 eV, equivalent to Q 10 of 2.80; Table 1; bootstrap hypothesis testing, p < 0.05). The Chl a does not significantly affect μ max (Eq. 2, t = À0.59, p = 0.56 > 0.05). Approximate E i estimated based on (sub)tropical dilution dataset The results of GAMs showed that the temperature, Chl a, and light were important factors affecting the phytoplankton growth rate in (sub)tropical waters (Fig. 4). After controlling the effects of Chl a and light, the E i 0 was 0.33 AE 0.06 eV, which equals a Q 10 of 1.56 and was also significantly lower than the temperature sensitivity within communities (E a = 0.80 AE 0.10 eV; Table 1; bootstrap hypothesis testing, p < 0.001).

Discussion
The thermal responses of the phytoplankton community are complicated because of the influence of community compositional changes, making it difficult to scale up the temperature-rate relationship from individual species to the whole communities. The theory derived from species levels may not directly guide the thermal responses of complex communities in nature. Here, we assessed the thermal response of phytoplankton communities using the data of short-term temperature manipulated experiments on natural phytoplankton communities, which incorporates the influence of community compositional shifts. Although the short-term temperature changes may induce the "thermal shock" to phytoplankton that may affect the estimate of temperature sensitivity, the value we estimated via nonlinear mixed-effect model (0.80 AE 0.10 eV) was in the range of literature reports on the within-species E a of all kinds of organisms including microbes, plants, and animals (Dell et al. 2011). It was also close to the average activation energy of 18 phytoplankton species (i.e., 0.77 eV) examined in a recent study (Barton and Yvon-Durocher 2019). Therefore, the average short-term temperature sensitivity (E a ) in the current study should be robust to reveal the emergent thermal response of phytoplankton communities, which is vital in highly thermal dynamic environments.
Our results demonstrated that the short-term temperature sensitivity of growth rate for a certain phytoplankton CV, the coefficients of variation; E a , the activation energy within the same community; E h , the deactivation energy induced by high-temperature inactivation; SE, the standard error of the estimate; T opt , the optimal temperature; μ c , the growth rate normalized to a reference temperature T c (15 C). Fig. 4. Partial effects of Boltzmann temperature (1/k b T c À 1/k b T), the natural logarithm of Chl a concentration (ln(Chl)), and the PAR on phytoplankton community growth rate based on a compiled dataset of dilution experiments conducted in the surface waters of (sub)tropical regions. Shaded areas denote 95% CI.
community (E a = 0.80 AE 0.10 eV; Table 1) was significantly higher than the temperature sensitivity of phytoplankton growth across communities at seasonal scale (E i = 0.51 AE 0.12 eV; Fig. 3). Moreover, it is also higher than the temperature sensitivity of phytoplankton across communities in the (sub)tropical waters (E i 0 = 0.33 AE 0.06 eV; Fig. 4), which were consistent with previous laboratory studies ( Barton and Yvon-Durocher 2019;Kontopoulos et al. 2020). Collectively, the results support the "partial compensation" hypothesis ( Fig. 1d) at the community level, suggesting that the effect of thermodynamic constraints on phytoplankton community growth could be only partially compensated. It questions the current monotonical temperature function of phytoplankton growth in biogeochemical models, which implicitly assumes that the temperature sensitivities within vs. across species are the same (Eppley 1972;Yool et al. 2013;Laufkötter et al. 2015).
The "partial compensation" pattern observed in phytoplankton communities reflects the seasonal succession of phytoplankton species (i.e., the succession of cold-adapted species in winter to warm-adapted species in summer). In winter, the phytoplankton community is dominated by cold-adapted species whose μ max at optimal temperature is constrained by low temperature. Thus, the μ max of winter community is significantly lower than the summer community (Fig. 3). This suggests that the thermal adaptation capacity of phytoplankton cannot fully compensate the thermodynamic constraint and is at least partly consistent with the "hotter is better" hypothesis which posits that organisms adapted to lower temperature have lower μ max (Fig. 1b;Frazier et al. 2006;Knies et al. 2009). When seasonal temperature changes, the phytoplankton community composition gradually shifts from cold-adapted species dominant to warm-adapted species dominant, leading to the increases in the T opt and μ max of the whole community. If there are only cold-adapted and warm-adapted species in the winter and summer communities, respectively, the E i across these communities may approach the E a within the community. However, the mixture of various species during seasonal succession may lower the μ max of the community, which eventually results in a relatively low E i . Indeed, some previous studies have found a low E i for plankton community metabolism and growth rates (L opez-Urrutia et al. Chen et al. 2012; Regaudie-De-Gioux and Duarte 2012). One possible explanation for these observations could be the community composition shift among various species with different thermal traits (e.g., cold-adapted and warm-adapted species). Nevertheless, the underlying compensatory mechanisms for thermal responses of phytoplankton at both species and community levels indeed remain substantially understudied and badly in need of thorough investigations.
Even so, such a "hotter is partially better" pattern we observed in the phytoplankton community has far-reaching implications for predicting the effect of climate warming on primary production and relevant processes (e.g., carbon export), especially for regional-scale predictions. Our results emphasized the difference in temperature sensitivity of phytoplankton growth within vs. across communities, as well as the difference between short-term and seasonal scales (i.e., E a vs. E i ). Such a difference has long been neglected in the biogeochemical and ecosystem models (Eppley 1972;Laufkötter et al. 2015). Applying the compensated temperature sensitivity to the regional models may not accurately capture the thermal response of phytoplankton community, especially in waters with large temperature fluctuations (Strock and Menden-Deuer 2021). The phytoplankton communities in such regions may not have sufficient time to adapt to the temperature changes. Thus, their thermal response could be stronger than the adapted communities. It has been found that the estimate of daily primary production in the highly thermal dynamic coastal waters is very different when using the short-term response of phytoplankton instead of the thermal response with acclimated effects (Strock and Menden-Deuer 2021). As the thermal fluctuations in coastal regions are anticipated to intensify under the context of climate changes (Schär et al. 2004), the fluctuations and the corresponding short-term responses of phytoplankton should be considered when predicting the effect of warming on primary production and relevant processes, which has been pointed out by a growing number of studies (Vasseur et al. 2014;Schaum et al. 2018). Nevertheless, using the short-term temperature sensitivity will neglect the effect of adaptation, which would overestimate the impact of ocean warming over timescales of years to decades. Thus, the compensated long-term temperature sensitivity should be used in the global models when predicting the long-time impacts.
In conclusion, to our knowledge, our study is the first to provide empirical evidence for the "partial compensation" pattern of the natural phytoplankton community, supporting the previous studies on phytoplankton species (Barton and Yvon-Durocher 2019;Kontopoulos et al. 2020). Our results emphasize the difference in temperature sensitivity at different timescales (short-term vs. seasonal scales) and highlight the importance of comparing the short-and long-term effects for predicting the ecosystem responses to ocean warming. We also suggest incorporating the short-term thermal response of phytoplankton into the regional models in highly thermal dynamic coastal ecosystems. Besides, it is worth noting that the phytoplankton growth rates in the current study were measured under the conditions of replete nutrient and simulated in situ surface light intensity, which should be sufficient for phytoplankton growth (Edwards et al. 2016). It allows us to study the thermal response of the phytoplankton without considering the complex effects of resource availability. However, resource availability usually affects the thermal response of the phytoplankton (Thomas et al. 2017;Liu et al. 2021), which entails further investigation of their interactive effects.