Adaptation potential of the copepod Eurytemora affinis to a future warmer Baltic Sea

Abstract To predict effects of global change on zooplankton populations, it is important to understand how present species adapt to temperature and how they respond to stressors interacting with temperature. Here, we ask if the calanoid copepod Eurytemora affinis from the Baltic Sea can adapt to future climate warming. Populations were sampled at sites with different temperatures. Full sibling families were reared in the laboratory and used in two common garden experiments (a) populations crossed over three temperature treatments 12, 17, and 22.5°C and (b) populations crossed over temperature in interaction with salinity and algae of different food quality. Genetic correlations of the full siblings’ development time were not different from zero between 12°C and the two higher temperatures 17 and 22.5°C, but positively correlated between 17 and 22.5°C. Hence, a population at 12°C is unlikely to adapt to warmer temperature, while a population at ≥17°C can adapt to an even higher temperature, that is, 22.5°C. In agreement with the genetic correlations, the population from the warmest site of origin had comparably shorter development time at high temperature than the populations from colder sites, that is, a cogradient variation. The population with the shortest development time at 22.5°C had in comparison lower survival on low quality food, illustrating a cost of short development time. Our results suggest that populations from warmer environments can at present indirectly adapt to a future warmer Baltic Sea, whereas populations from colder areas show reduced adaptation potential to high temperatures, simply because they experience an environment that is too cold.

populations that otherwise would go extinct or be in need of migration to colder areas (Davis & Shaw, 2001;Foden et al., 2013;Hughes et al., 2003). Evidence exists for some populations that have adapted to high temperatures (Lonsdale & Levinton, 1985;Yampolsky, Schaer, & Ebert, 2014), and if species exhibit adaptations at present, it is likely that they will in the future as well (Merilä & Hendry, 2014;Stoks, Geerts, & De Meester, 2014). However, there are still important topics to address on how species may adapt to climate change, such as how contemporary populations can adapt to future conditions. This can be estimated by quantifying indirect selection between present and future environments, which is revealed by the sign and strength of genetic correlations. Moreover, many studies focus on one factor at a time (Todgham & Stillman, 2013), and hence, much less is known about the effect of multiple factors interacting simultaneously with temperature (Stoks et al., 2014).
Adaptive potential is essentially genetic variance (Foden et al., 2013;Urban et al., 2014), from which a series of estimates related to selection and adaptation can be calculated. For example, if a trait is measured on groups of full siblings, the proportion of phenotypic variance that is caused by between-group variance is the heritability, which is a predictor between direct selection and adaptation. More so, if the same trait is measured across different environments, the correlation of between-group variances in each environment is the genetic correlation, which is a predictor between indirect selection and adaptation. Indirect selection depends on the sign of the correlation and may either reinforce, antagonize, or have no effect on adaptation (Etterson & Shaw, 2001). For example, if a high value of the same trait is of benefit in two environments, such as extant and future conditions, and the genetic correlation between the trait in both environments is positive, then selection at extant conditions will render a high value also in future conditions through indirect selection. Hence, genetic correlations are highly relevant for inferences of local adaptation and for the adaptation potential of populations.
The difference in trait value between two or more environments is the phenotypic plasticity; this is an environmentally induced change in the phenotype that enables a single genotype to respond differently to various environmental conditions (Via et al., 1995). Plasticity may also vary between genotypes in response to the environment, that is, an interaction between the genotype and the environment (Falconer & Mackay, 1996;Lee, 2002;Saltz et al., 2018). The variance between genotypes in different environments may reveal if selection in one environment will have a correlated, indirect, response in another environment. Hence, there is a formal link between the genotype by environment interaction and the genetic correlation (Falconer, 1990;Falconer & Mackay, 1996).
For zooplankton, development time is a useful trait for studying adaptation since it is intimately connected to fitness, with a shorter development time increasing the exponential fitness parameter r and hence population growth (Allan, 1976;Lewontin, 1965). Species in seasonal environments that produce several generations over the year, should in theory, benefit if the development time is as short as possible when conditions are favorable (Allan, 1976;Kingsolver & Huey, 2008;Roff, 1980). Body size and fecundity are also important for population growth rates of zooplankton; however, they are relatively less important than the time lag between generations (Allan, 1976).
Typically, populations with a short development time are comparably smaller when they reach maturity than populations with longer development time (Kingsolver, Massie, Ragland, & Smith, 2007;Merilä, Laurila, & Lindgren, 2004;Sniegula, Golab, Drobniak, & Johansson, 2016). Hence, a fitness trade-off between size (via fecundity) and development may influence the evolution of thermal reaction norms. Although, exceptions from the typical trade-off exists where populations can maintain both fast development and large size at maturity (Gotthard, Nylin, & Wiklund, 1994;Tang, He, Chen, Fu, & Xue, 2017). However, maximizing both traits involves increased growth rates and can result in higher susceptibility to starvation (Gotthard et al., 1994;Stoks, Block, & McPeek, 2006). Thus, overcoming one trade-off includes a new trade-off. This is important in a scenario where other stressors may change in addition to temperature and indirectly affect organisms' response to temperature.
The calanoid copepod Eurytemora affinis is at places one of the dominating zooplankton species in terms of number and mass in both freshwater and coastal estuaries, and hence an important grazer and prey for plankton feeding fish (Diekmann, Clemmesen, John, Paulsen, & Peck, 2012;Hernroth & Ackefors, 1979;Rajasilta, Hänninen, & Vuorinen, 2014). In the Baltic Sea, E. affinis forms large transitory populations that typically peak in late summer (Hernroth & Ackefors, 1979). Given this opportunistic (r) life strategy, it is expected that E. affinis has a development time that is as short as physiologically possible when conditions are favorable. Eurytemora affinis consists of a species complex with a widespread distribution in the northern Hemisphere (Lee, 2016). Within the complex, both development time and body size differ between populations (Karlsson, Puiac, & Winder, 2018;. More so, the populations are highly variable in diverse traits, such as morphology, habitat use, ecological effects, and salinity tolerance (Favier & Winkler, 2014;Lee, Remfert, & Gelembiuk, 2003). Clades and lineages are also spread outside their native range because of maritime traffic and introduced into other environments (Sukhikh, Souissi, Souissi, & Alekseev, 2013;Winkler, Souissi, Poux, & Castric, 2011). However, the rapid adaptations recorded in this species complex support that even invasive populations might be locally adapted to their new environments (Lee, 2002;Lee, Posavi, & Charmantier, 2012;Lee, Remfert, & Chang, 2007).
The Baltic Sea is one of the marine areas with the highest recorded temperature increase during the past century (Meier, 2015), and climate change may increase precipitation in the catchment area possibly leading to lower salinity and changes in food web structure (Lefebure et al., 2013;Meier, 2015). The Baltic Sea spans over a large latitudinal and ecological gradient and consists of different basins that vary in temperature, salinity, and food web structure (i.e., trophic states, terrestrial organic matter) (Andersen et al., 2017;Larsson, Elmgren, & Wulff, 1985;Lefebure et al., 2013;Lehmann, Getzlaff, & Harlaß, 2011). The copepod E. affinis is widely distributed in the Baltic Sea, and populations are thus subjected to different environmental conditions and to different selection pressures depending on their geographical position.
The aim in this study was to investigate if the copepod E. affinis may adapt to a future warmer Baltic Sea. For this, a quantitative genetics approach was used, with related individuals (full siblings) crossed over different temperatures in common garden experiments. Eurytemora affinis was further exposed to different temperatures in combination with different salinity and food type to explore interactions of multiple stressors. For this, populations that originate from areas of different temperature, salinity, and primary production were compared to investigate local adaptations and trade-offs.

| Study populations and rearing conditions
Eurytemora affinis were collected with 90 µm vertical tow nets in autumn 2014 from the Bothnian Bay (BB, monitoring station F3A5, 65°10.14', 23°14.41'), the Gulf of Riga-Pärnu Bay (GOR, 58°21.67', 24°30.83'), and the Stockholm Archipelago-Askö (STHLM, monitoring station B1, 58°48.19', 17°37.52'). The GOR population has in previous studies shown to develop to adult faster and at a larger size ( Figure 1) than the STHLM population . Copepods were transported to the department in cooled conditions and placed in a cold room where temperature gradually increased up to 17°C over the course of several days.
In the laboratory, a minimum of 300 individuals were sorted out from each location and put into cultures maintained at 17°C and salinity of seven practical salinity units (PSU, g/kg). Tap water was used for the stock cultures and breeding, and the water was circulated in an aquarium for approximately 1 week with gravel from a small stream, making it more habitable for aquatic organisms. The water was then mixed with Instant Ocean TM to reach appropriate salinity. The copepod cultures were fed two types of Cryptophytes: Rhodomonas salina and Rhinomonas nottbecki. The copepods were reared at a relatively high temperature, 17°C, at which E. affinis reproduces relatively rapidly and could undergo many generations at common conditions. Before the experiments, all populations had gone through at least three generations, likely many more, in common conditions in order to control for environmental and maternal effects (Sanford & Kelly, 2011). The choice of salinity was based on the survival of the food source, R. salina, which did not grow well at lower salinities.

| Analyses of environmental conditions
Environmental data on temperature, salinity, and chlorophyll-a were obtained from the Swedish Meteorological and Hydrological Institute (SMHI) for the BB (station F3) and STHLM (station B1) sites and the International Council for the Exploration of the Sea (ICES) for the GOR site. The GOR population was not sampled at a monitoring station, and hence, data are from the geographical cut-off: highest lat, lon 58°35.00', 24°47.17'; lowest 58°02.50', 24°17.17'. All available observations from depth ≤10 m between the years 1993 and 2018 were used for the analyses. The data were analyzed as nine separate generalized additive models (GAM), one for each population and parameter combination, and a smooth function was applied to the linear predictor day of year.
The GAM models were fitted by the use of package mgcv (Wood, 2004(Wood, , 2011. The predicted fitted values and 95% CI were used to assess any "significant" differences in parameters between sites.  addition, monitoring data on Eurytemora sp. abundance from SMHI and ICES were analyzed with GAMs in order to visualize timing of population abundance peaks over the same time period as for the environmental data. Here, abundances of different life stages from each sample were added up and predicted over day of year.

| Common garden experiments
Two common garden experiments were designed; the first experiment took place in April-June 2015 and the second in January- salina was used as food.
For the second experiment, two populations GOR and STHLM were used, and in total 28 families and 283 individuals that matured to adults (Table 1). Here, two temperatures 17 and 22.5°C, two salinities 2 and 7 PSU, and two types of food Cryptomonas sp. and R.
nottbecki were used. Both food type and salinity were crossed over temperature and population; however, food and salinity were not fully factorial because Cryptomonas sp. could be cultured in 2 but not 7 PSU. In contrast, R. nottbecki was cultured in salinity 2 and 7 PSU and was therefore used as food in both salinity treatments. In comparison with R. salina from the first experiment, R. nottbecki is in size (c. 12 µm long and 5 µm wide), shape, growth rate, and color very similar (personal observation) and we assumed they are of similar and high food quality. Cryptomonas sp. is slightly bigger than the other two species (c. 20 µm long and 10 µm wide). All three species are members of the phylum Cryptophyta, R. salina and R. nottbecki belong to the family Pyrenomonadaceae, while Cryptomonas sp. belongs to Cryptomonadaceae. Cryptophytes are in general known as a good food sources for calanoid copepods leading to a short development time and high egg production (Knuckey, Semmens, Mayer, & Rimmer, 2005;Koski, Breteler, & Schogt, 1998).
To obtain full sibling clutches, E. affinis mature males and copepodite females (that would later undergo sexual maturity) were paired up in 15 ml cylinders at 17°C, and this procedure ensured that only one male fertilized the eggs as copepod females may store sperm (Allan, 1976). Once the egg sacks became visible, the eggs were separated with an injection needle under a stereomicroscope and placed into 10 ml vials, with 1-3 eggs per vial depending on clutch size. Eggs from each full sibling clutch (family) were split across temperature (experiment one), as well as temperature*food (experiment two) and temperature*salinity (experiment two) with two vials for each family and treatment combination. Thereby, family lines were put in specific environments, which make it possible to separate between genetic and environmental variance.
For the experiments, the aquarium water was filtered through a 0.7 µm pore size filter (WhatmanTM GF/F) before adding food and copepod eggs to the vials. The algae were observed every day to ensure that they remained in a healthy state during the experiment, which is reflected in the color of the water and is pink-red for R.
salina and R. nottbecki and green for Cryptomonas sp. In some vials, the algae culture died and was replaced as soon as it was detected.
The algae suspension in the experimental vials had a concentration of approximately 200,000 cells per ml, and this concentration is well above ad libitum for E. affinis (Ban, 1994). The vials were put in racks in temperature incubators (INKU-line from WVR) with a precision of ±0.5°C.
Development time from nauplii (newly hatched) to adult and survival from nauplii to adult were the two response parameters, and the explanatory variables were temperature, food type, and salinity.
Copepods undergo six nauplii and six copepodite stages where the sixth stage is the adult. Once per day, the number of living individuals and their life stage was observed. Individuals were classified as adults when females developed spike like extensions at the end of their prosome (one on each side of the urosome) and a distinct furca, males when they developed wavelike antennas and a distinct long furca (Katona, 1971).

| Statistical analyses of life-history traits
All analyses of data were done with R (R Core Team, 2019) and all figures by using the R package ggplot2 (Wickham, 2009). Development time and survival were analyzed in mixed models, functions lmer and glmer from the lme4 package (Bates, Mächler, Bolker, & Walker, 2015). Response variables were Gaussian for development time and binomial for survival. Fixed factorial effects for the models were the interaction of population and the experimental treatments, and family line was used as random effect. Treatment effects were analyzed as factors; thus, each factor combination represents a character state (Ghalambor, McKay, Carroll, & Reznick, 2007). Mixed model outputs were analyzed with type two ANOVAs using the car package (Fox & Weisberg, 2011). From the mixed models, a selection of contrasts between treatment combinations and associated p-values are presented in the results. For contrasts of development time, the mixed model was fitted with function lme from the nlme package (Pinheiro, Bates, DebRoy, & Sarkar, 2017).
In the second experiment, the setup was not fully factorial, because the food type Cryptomonas sp. could not survive at 7 PSU, and hence, this treatment combination does not exist, and the interaction between population*temperature*salinity*food could not be tested. Therefore, the data were split in two analyses, one for population*temperature*salinity and one for population*temperature*food. The reason for not including both three-way interactions in one model was that some factors would average over the uneven treatment. For example, the effect of salinity would compare the average of the two food types at 2 PSU with only one food type at 7 PSU. By dividing the data set into one for salinity and one for food type, two analyses of the main effects temperature and population, and the temperature*population interaction are presented in the results. However, both analyses led to the same conclusions, and both are presented in the results.

| Broad sense heritability, genetic correlation, and interaction of genotype and environment
Heritability is a measure of the degree of resemblance between relatives; it aims to predict the phenotype of progeny from the phenotype of parents. In the context of heritability, an individual has two values, the phenotypic value, that is, the measured metric character, and the breeding value, that is, the average phenotype of its progeny expressed as deviations from the population mean (Falconer & Mackay, 1996). The phenotypic value is observable, but the breeding value is unobservable for the individual. The heritability provides a link from the selected phenotypes to the phenotype of the next generation. Hence, for selection and adaptive potential, the change in mean phenotype of a population has to be predicted from the correspondence between the parent phenotype and offspring. This is done by the breeder's equation: R = h 2 × S, where R is the response to selection, h 2 is the heritability and S the difference from the population mean to the mean of the selected individuals (Falconer & Mackay, 1996, eq. 11.2). The heritability is for a full sibling design estimated from the intraclass correlation: is the between-group variation and 2 is the Gaussian residual error variance, the heritability is then t ≥ 0.5 h 2 (Falconer & Mackay, 1996, table 10.2), where 0.5 is the average relatedness of full siblings.
The genetic correlation is similar to the heritability in the way that it estimates the link between phenotypic values and breeding values. However, here, the phenotypic value in one trait predicts the breeding value of the other trait. In the present study, the full siblings are crossed over temperature, and hence, it is possible to estimate the correlation of development time at different temperatures. Falconer (1952) and Yamada (1962) proposed that the same trait when measured in a different environment can be regarded as a different trait. This is because the physiology of the organism is expected to be different depending on environment and consequently also the genes required differ between the environments (Falconer & Mackay, 1996). The calculation of the correlation of the same trait at different temperatures is analogous to that of heritability as it is the correlation of between-group variances at each temperature (Falconer & Mackay, 1996, eq. 19.2), where COV is the covariance of the families between two different temperatures (X and Y), and 2 is the between-group variance of the families at a specific temperature (X or Y). The correlated response to selection where i the intensity of selection, h X and h Y are the heritability in the two environments, G r the genetic correlation, and PY the standard deviation of the phenotypic value for character Y (Falconer & Mackay, 1996, eq. 19.6). The correlated, indirect, response to selection is weaker than direct selection; the two can be compared by (Falconer & Mackay, 1996, eq. 19.9).
The genotype by environment interaction and the genetic correlation are related in such a way that a specific configuration of reaction norms will lead to a specific correlation (Falconer, 1990).
The genotype by environment interaction estimates the performance of each genotype, that is, family, from one environment to the next, and is as any interaction, a test of differences in slopes (Saltz et al., 2018). The variance of the family differences from the average reaction norm is the between-group variance and creates a formal link between the interaction and the correlation (see results on genotype by environment interaction). For local adaptation, both estimates are fundamental as they describe how much of a phenotype is carried over from selection in one environment to its progeny in the next environment. In the present study, a short development time is assumed to be the best performance, and hence, a positive correlation between two environments would indicate that the best genotype in one environment also is best in the other environment.
A negative or low correlation would indicate local adaptation and that selection has to be carried out in each environment separately to achieve the best performance.
Genetic correlations and broad sense heritability of development time were estimated by MCMC sampling using the function MCMCglmm (Hadfield, 2010). For genetic correlations and heritability, the unit of replication is at the family level; therefore, the data from both experiments were pooled to increase the precision of estimates. A very large number of replicates on family level are needed for any precise estimates of heritability and genetic correlations; this is typically not feasible in experimental studies and is instead more often available in animal breeding (Hoffmann, Merilä, & Kristensen, 2016). Nevertheless, an optimal design for heritability should reduce family size on behalf of a higher number of families.
The optimal design is achieved when the sampling variance of the intraclass correlation is minimal, which it is when n = 1/t (Falconer & Mackay, 1996, chapter 10). However, t is not known before the experiment starts, and in the present study, the theoretical optimal family size was 1/t = 5.7, and the actual family size was on average 556/65 = 8.6 for the complete data set. Including larger families than the optimum is preferable as it is difficult to predict the percentage of individuals that will develop into adults beforehand, and hence, the resulting family size. For the correlation and heritability models, fixed effect covariates were included to control for the variance caused by the treatments and to avoid confounding effects on the between-group variance and error variance (Nakagawa & Schielzeth, 2010). The models contained the following covariates: population, temperature, salinity, and food, when there was more than one treatment level per covariate.
For the MCMC models, inverse-Wishart priors for the random effect were used; for heritability the variance was set to 2 and the belief parameter to 1 for the G-structure (group), for the R-structure (residual), respective values were 1 and 0.002. The belief parameter sets the values of the model parameters and describes the shape of the prior distribution. In the context of a mixed model, a group contains observations that are not independent, that is, the different full sibling families make up unique groups. For genetic correlations, the prior variances were set as the true variance for each trait (development time at a specific temperature) and the belief parameter to 3 (i.e., n dimensions of the G matrix + 1) (Hadfield, 2019;Wilson et al., 2010). The models ran for 2.6 million iterations with a burn-in of 600,000 and sampled every 1,000 iteration, which generated an effective sample size of 2,000. From the 2,000 samples, the median and 95% quantiles (0.025, 0.975) are presented for heritability, and for genetic correlations, the mode and the 95% highest posterior density are presented.
The significance of the genotype by environment interactions was tested by model comparison in an analysis of deviance. One model with the temperature + family was compared with a model with the additional temperature*family term. The models were simple linear regressions with Gaussian error distribution; significance was assessed by F-ratio tests. Furthermore, the variances of the fixed effect temperature and the random effect family across temperature (temperature|family) were quantified and compared by linear mixed models (lme). The analysis of deviance tests whether the reaction norm slopes are different for the families, whereas the mixed models quantify the variances of the overall effect of temperature and the variance of families across temperature (Bolker et al., 2009). Thereby, both the magnitude and the significance of the genotype by environment interaction were compared across temperature. The genotype by environment interactions was estimated for the same set of conditions as for the genetic correlations. The highest increase in temperature is predicted in the northern Baltic Sea (Meier, 2015), in year 2069-2098; this will result in similar maximum summer temperatures between BB and STHLM, 19.7 and 19.5°C, respectively, but temperature will remain the highest in GOR, 22.5°C (Figure 2).

| Survival
In the first experiment (Table 2; Figure 3b), there was no significant main effect of temperature on survival, but a significant interaction of population and temperature ( 2 4 = 24.28, p < .001). Furthermore, survival was in general lower for the BB population compared with the GOR and STHLM populations ( 2 2 = 38.22, p < .001).
In the second experiment (  R. nottbecki with respective 40% (24, 58) and 88% (73, 96) survival (z = 3.85, p < .001), while at 17°C, there was no difference between food types. For the STHLM population, the type of food had no effect, and hence, the temperature*food interaction was mainly driven by the GOR population. Furthermore, the main effect of salinity on survival was significant ( 2 1 = 5.53, p = .019), and survival was higher at 2 PSU where it was 76% (66, 83) than at 7 PSU where it was 62% (51, 71).

| Genotype by environment interaction, genetic correlations, and broad sense heritability
Genotype by environment interaction was significant between 12 and 17°C, and 12 and 22.5°C but not significant between 17 and 22.5°C (Table 3, Figure 5). The results from the analysis of deviance were in agreement with the results from the linear mixed models. F I G U R E 4 Development time from nauplii to adult (a, c) and survival from nauplii to adult (b, d) across temperature for the two Eurytemora affinis populations, GOR in blue triangles and STHLM in orange squares. The two upper panels show the interactive effect of population*temperature*salinity, and the dashed line shows 7 PSU and the solid line 2 PSU. The two lower panels show the interactive effect of population*temperature*food, and the dotted line shows Cryptomonas sp. and the solid line R. nottbecki. Estimates and 95% confidence intervals are from the model output. Note that the estimates for salinity 2 PSU and the food R. nottbecki are identical in (a, c) and (b, d), see methods

TA B L E 3
Analysis of deviance output with type II sums-of-squares for genotype by environment interactions. Models with and without the interaction term were compared between the temperatures stated in the "model" column. p-values were calculated as F-ratio tests on the difference in deviance and degrees of freedom between models That is, when the genotype by environment interactions were significant, the variance of the interaction was also greater, and hence, there was more variation in phenotypic plasticity (Figure 6a). The variance in phenotypic plasticity was greater between the coldest temperature 12°C and the two higher temperatures (17 and 22.5°C), than between the two higher temperatures (Figure 6a).
Genetic correlations between temperature treatments were not significantly different from zero between 12 and 17°C and between

| D ISCUSS I ON
This study explores the selection and adaptation potential to changing environmental conditions of the copepod E. affinis, a key zooplankton species in coastal waters and in the Baltic Sea. We found , and (f) show the same data as the panels above, but here as correlations between temperatures of the mean centered family averages with regression lines. The variances of the mean centered family averages are the between-group variances that were used to estimate heritability and genetic correlations. The number of families and individuals in each panel is as in Figure 6b for the same correlation/reaction norm E. affinis to be adapted to different temperature regimes and that the species can adapt to higher temperature than present via indirect selection at 17°C, which can result in an adaptation at 22.5°C.
However, our results suggest that the adaptation to high temperature is unlikely to occur for populations located in "colder" temperatures, that is, 12°C. Global warming coupled with changes in food conditions and salinity may alter temperature tolerance, and the benefits of temperature adaptations may be compromised if additional changes in salinity and food conditions co-occur.
Our results indicate that selection at a present temperature can facilitate adaptation to a more extreme future temperature. This because family lines that perform well at intermediate temperature will also perform well at higher temperature, indicated by the positive genetic correlation between 17 and 22.5°C, which confirm indirect selection, and hence, adaptive potential between the two temperatures ( Figures 5 and 6b). The reaction norms of the genotype by environment interaction, E|G, between 17 and 22.5°C had in comparison lower variances than between the other temperatures ( Figures 5 and 6a), indicative of overall low variance in phenotypic plasticity. Although low variance in phenotypic plasticity is typically seen as a limit of the evolutionary response (Dam, 2013;Ghalambor et al., 2007;Lee, 2002;Oostra, Saastamoinen, Zwaan, & Wheat, 2018;Sgrò, Terblanche, & Hoffmann, 2016), it is possible to see its potential benefits because all genotypes are more prone to respond similarly to both direct and indirect selection, and a short development time is likely beneficial at both 17 and 22.5°C. Hence, the interaction is not adaptation potential per se, as a significant genotype by environment interaction can result in antagonising selection as well. Therefore, the configuration of reaction norms, which determines the sign and strength of the genetic correlation, should preferably be considered together with the genotype by environment interaction to assess adaptive potential. In contrast, variance in the reaction norms between the cold (12°C) and the highest temperature (22.5°C) was greater and the genetic correlations indicated that indirect selection of development time at 12-22.5°C is unlikely ( Figures 5 and 6ab). Hence, selection on a genotype with a shorter development time compared with the population mean at 12°C will likely have no effect on the development time at 22.5°C.
The populations had different development time at the highest temperature treatment with the GOR population having the shortest, STHLM intermediate, and BB the longest. Long and warm summers create better opportunities for adaptation to warm temperatures. The warm summer season is the time when abundances are the highest and consequently genotypes compete via their population rate of increase, and hence, they benefit by having as short generation times as possible. The development time in the present study was ordered as GOR<STHLM<BB, and temperature and chlorophyll-a from the sites are ordered as GOR>STHLM>BB. For zooplankton, higher temperatures and more food lead to a shorter development time (Ban, 1994;Gillooly, 2000). Thus, the population from high temperature and food availability had a shorter intrinsic development time compared with the populations originating from lower temperatures and poorer food conditions. Hence, the covariance of the populations' environmental values and the populations' genotypic values is positive and, therefore, indicative of a cogradient variation (Conover, Duffy, & Hice, 2009;Falconer, 1990). In addition, the results suggest that life in a cold environment constrain evolution of increased performance in a warm environment, that is, F I G U R E 6 Panel (a) shows the environmental (E) and genotype by environment (E|G) variances expressed as standard deviations for the reaction norms between temperatures. Stars show significance of the genotype by environment interaction presented in Table 3 warm adaptation (Angilletta, Huey, & Frazier, 2009;Frazier, Huey, & Berrigan, 2006 (Lang, Hodac, Friedl, & Feussner, 2011) and between freshwater and marine species (Galloway & Winder, 2015), where freshwater species, such as Cryptomonas sp., tend to have lower quality compared with marine species. At high temperature, the GOR population develops to maturity in a shorter time and to a larger size than the STHLM population . The combination of large size at maturity and short development time is unusual among ectotherms. Compared between populations, more often a tradeoff of these two traits exists, where fast development comes with small size (Allan, 1976;Gillooly, Charnov, West, Savage, & Brown, 2002;Kingsolver & Huey, 2008;Merilä et al., 2004;Roff, 1980 (Gotthard et al., 1994;Stoks et al., 2006 Lee et al., 2003). However, decreased salinity affects metabolic rates and ingestion rates of E. affinis, and freshwater tolerance increases if the copepods are exposed to sufficient food availability, as in our experiment (Hammock et al., 2016;Lee et al., 2013).
Increased feeding rates may thus both shorten development time and increase freshwater tolerance (Ban, 1994;Lee et al., 2013). It is therefore possible that the lower salinity evokes a stress response that leads to increased feeding, which in turn leads to shorter development time in the GOR and STHLM populations when salinity was reduced from 7 to 2 PSU (Figure 4a).
The heritability of E. affinis development time calculated from the complete data set was estimated to be 0.35. Heritability is typically low for life-history traits that have high impact on fitness, such as development time (Berger, Postma, Blanckenhorn, & Walters, 2013;Bradshaw, Holzapfel, Kleckner, & Hard, 1997;Sniegula et al., 2016), and gives a direct measure of how much development time can change from one generation to the next. It is difficult to predict whether adaptation will take place within populations or if populations can reproduce and mix, forming metapopulations on which selection can act on. However, as E. affinis consists of a cryptic species complex with distinct populations that may be reproductively isolated even when they are co-occurring (Favier & Winkler, 2014;Lee, 2000), it implies that it is uncertain if adaptations can happen by crossings of populations from warm and cold environments. North American east coast have been found to be reproductively isolated (Lee, 2000); adaptation to temperature, salinity, and food conditions in E. affinis is likely limited within cryptic species that inhabit different environments.

| SUMMARY
Our study shows that selection of development time at warmer temperatures of 17 and 22.5°C is positively correlated, and hence, E. affinis can adapt to higher temperatures if they currently inhabit waters of ≥17°C because of indirect selection that reinforce adaptation to high temperatures. In contrast, selection at cold and warm temperature was uncorrelated, and a population at 12°C is unlikely to adapt to 22.5°C. In agreement with the sign of the genetic correlations, we found that the population from the warmest site of origin

CO N FLI C T O F I NTE R E S T
The authors declare no conflicts of interest.

AUTH O R CO NTR I B UTI O N S
MW and KK: designing and writing. KK: performing the experiments, maintaining cultures of organisms, data analysis, and writing (first draft).