You Better Repeat It : Complex CO 2 × Temperature Effects in Atlantic Silverside Offspring Revealed by Serial Experimentation

Concurrent ocean warming and acidification demand experimental approaches that assess biological sensitivities to combined effects of these potential stressors. Here, we summarize five CO2 × temperature experiments on wild Atlantic silverside, Menidia menidia, offspring that were reared under factorial combinations of CO2 (nominal: 400, 2200, 4000, and 6000 μatm) and temperature (17, 20, 24, and 28 ◦C) to quantify the temperature-dependence of CO2 effects in early life growth and survival. Across experiments and temperature treatments, we found few significant CO2 effects on response traits. Survival effects were limited to a single experiment, where elevated CO2 exposure reduced embryo survival at 17 and 24 ◦C. Hatch length displayed CO2 × temperature interactions due largely to reduced hatch size at 24 ◦C in one experiment but increased length at 28 ◦C in another. We found no overall influence of CO2 on larval growth or survival to 9, 10, 15 and 13–22 days post-hatch, at 28, 24, 20, and 17 ◦C, respectively. Importantly, exposure to cooler (17 ◦C) and warmer (28 ◦C) than optimal rearing temperatures (24 ◦C) in this species did not appear to increase CO2 sensitivity. Repeated experimentation documented substantial interand intra-experiment variability, highlighting the need for experimental replication to more robustly constrain inherently variable responses. Taken together, these results demonstrate that the early life stages of this ecologically important forage fish appear largely tolerate to even extreme levels of CO2 across a broad thermal regime.


Introduction
The current anthropogenic increase in atmospheric and therefore oceanic carbon dioxide (CO 2 ) concentrations has been unparalleled over the past 66 million years [1].Resultant changes in ocean pH and carbon chemistry (ocean acidification, OA) are likely to have major impacts on marine ecosystems [2] by changing species abundances, interactions and trophic dynamics, all of which depend ultimately on the CO 2 sensitivities of individual organisms [3][4][5].Laboratory experiments have played an important role in quantifying these CO 2 sensitivities, suggesting that they are greater in sessile, calcifying invertebrates than in active, non-calcifying vertebrates, and greater in early life stages than adults [6][7][8].The latter has been particularly well documented for marine fish, where adults are largely tolerant of acute high-CO 2 levels far exceeding predicted OA conditions [9,10].By contrast, fish early life-stages (embryos and early larvae) that are still developing effective acid-base regulation have exhibited reduced survival [11,12], reduced growth [13,14], defective development [14,15], otolith over-calcification [16,17], and behavioral abnormalities in response to high-CO 2 conditions in the laboratory [18,19].Experiments showing no discernible CO 2 effects are also common [20][21][22][23][24].This complexity of empirical evidence remains challenging to reconcile [25], but is consistent with the emerging consensus of species-and population-specific CO 2 sensitivities, particularly for fish adapted to high CO 2 and pH variability in their habitats [26].
To date, experimental approaches have largely been guided by open-ocean predictions for administering CO 2 treatments (see [27]).It is now recognized, however, that many marine organisms experience considerable diel and seasonal pH/CO 2 fluctuations in their habitats [26,[28][29][30].Short-term pH/CO 2 variability can be attributed to ephemeral upwelling [31], river input [32], and metabolic processes that dominate CO 2 variability in coastal habitats [33] and in oxygen minimum zones [34].The seasonal intensification of community respiration in highly productive coastal systems (e.g., saltmarshes and mangrove lagoons) can increase both average and extreme CO 2 levels to nearly double the open-ocean OA predictions for the next 300 years [35].Given the thermal sensitivity of microbial respiration rates, metabolically driven acidification is generally most extreme during peak summer temperatures [36].Hence, to better understand climate change effects on coastal species, experiments should implement CO 2 and temperature conditions that reflect the range of modern and predicted conditions of their source ecosystems, rather than relying on average global predictions.
While single-factor CO 2 experiments are a necessary initial step, it is now widely recognized that OA proceeds in concert with ocean warming and deoxygenation.Experiments are needed to address species sensitivities to multiple stressors of marine climate change [37,38].Warming may be the primary driver of ecological disruption, as there is already evidence of shifting fish distributions and phenologies [39,40], which likely reflect the need for ectotherms to maintain environments within their scope of physiological optima [41].The capacity of organisms to maintain performance at temperatures approaching or exceeding their thermal tolerance is a key metric in determining climate sensitivity [42].Elevated environmental CO 2 may increase energetic costs associated with acid-base regulation [43] and could compromise the functional capacity of other vital processes [44] and therefore increase an organisms' sensitivity to thermal extremes [38].Thus, CO 2 × temperature experiments are not only more realistic, they may also discover important stressor interactions that elude single-stressor approaches [45].
The majority of studies evaluating CO 2 × temperature effects in fish have focused on stenothermal taxa from polar [46][47][48][49][50] or tropical habitats [51][52][53][54].These fish are presumably adapted to their relatively stable thermal environments and may thus show limited acclimation capacity to combined climate stressors [55,56].By contrast, temperate species are often eurythermal, i.e., capable of acclimating to broad seasonal temperature fluctuations.However, they are still adapted to specific thermal regimes [41] and often show narrower thermal requirements during CO 2 sensitive early life stages [14,57].
Many fitness-relevant traits such as growth or survival are highly variable in nature during fish early life stages, thus producing variable outcomes even under most meticulously controlled experimental conditions [58,59].Variations in offspring due to parentage, food quality and quantity, or water sources can introduce additional variability, hence underscoring the risks of generalizing results from single experiments to population or species characteristic such as CO 2 or temperature sensitivity.More robust depictions of CO 2 and temperature sensitivity are likely to emerge if experiments are replicated and analyzed together, but this approach is still underutilized in studies of climate change effects on marine organisms.
Here we report on five factorial CO 2 × temperature experiments conducted on offspring of wild Atlantic silversides, Menidia menidia, an ecologically important and abundant coastal forage fish with a broad distribution along the east coast of North America [60].Wild silverside offspring are amenable to experimental manipulations and have thus become a widely used model in OA experiments [58,[61][62][63][64][65][66].Over the course of three years, we repeatedly reared Atlantic silverside offspring at different factorial combinations of CO 2 and temperature to quantify the temperature-dependence of CO 2 effects in growth and survival.We hypothesized that negative responses to high-CO 2 levels would largely occur at the species lower and upper thermal limits, while predicting fewer or no CO 2 effects at optimal thermal conditions.

Field Sampling and Experimental Designs
Collections of wild, spawning ripe Atlantic silverside were made during high tide 1-3 days prior to full or new moons following the species' semi-lunar spawning periodicity during spring and early summer [61].Adults were caught with a 30 × 2 m beach seine from local salt marshes and transported live to our laboratory facilities.For the 2014 experiment (experiment 1), adults were collected from Poquot Beach (40    Strip-spawning protocols maximized fertilization success, while enabling random distribution of embryos across replicates [61,63].For each experiment, eggs from 12+ running-ripe females were gently mixed into shallow plastic dishes lined with 1-mm plastic window screening.Milt from each of 20+ males was collected and pooled into 500-mL glass beakers, mixed with seawater, stirred, then gently poured into spawning dishes and mixed with eggs for ~15 min.The number of spawners used for each experiment and their length measurements are reported in Table S1.In this species, fertilized embryos uncoil chorionic filaments, which readily attach to screening.Window screens were cut into smaller sections where embryos were counted under low magnification with high accuracy.Experiments were initiated within 2 h of fertilization when replicate rearing-containers (20-L cylindrical polyethylene buckets) received precisely 100 embryos.Rearing-containers were filled with clean seawater (filtered to 1 µm and UV sterilized).Optimal salinity (27)(28)(29)(30)(31) and light conditions (15 h light:9 h dark) for rearing M. menidia were maintained across experiments [60].The number of CO 2 × temperature treatments and replicates varied between experiments (see Table 1).For actual CO 2 × temperature treatments administered see Table 2. Starting four days post-fertilization (dpf), each rearing-container was checked for hatched larvae.On the morning of first observed hatch, larvae were immediately provided with equal rations of powdered weaning diet (Otohime Marine Fish Diet, size A1, Reed Mariculture ® , Campbell, CA, USA) to stimulate feeding and ad libitum levels of newly hatched brine shrimp nauplii (Artemia salina, San Francisco strain, brineshrimpdirect.com,Ogden, UT, USA).Larvae were fed daily ad libitum rations of newly hatched nauplii for the remainder of the experiment.To quantify survival to hatch, one day post-hatch larvae were counted by gently scooping small groups into replacement rearing-containers.For initial hatch standard length (SL, nearest 0.01 mm) measurements, larvae were randomly sub-sampled (N = 10) from each replicate were preserved in 5% formaldehyde/freshwater solution buffered with saturated sodium tetraborate.The timing of hatch sub-samples varied slightly between experiments and temperatures (see Table 3).Rearing-containers were siphoned of waste daily, and treatment water was partially exchanged with new seawater every other day.Levels of ammonia waste were monitored daily (Saltwater Ammonia Test Kit, API ® , Chalfont, PA, USA) to maintain uncritical levels below 0.25 ppm.All experiments were terminated when larvae reached ~10 mm SL within temperature treatments (determined by visual estimates).Using body size rather than set time intervals allowed comparing CO 2 effects on offspring during the same developmental period (i.e., fertilization to ~10 mm SL) across temperature treatments.Experiment durations ranged from 14 to 36 days (Table 3).At termination, all survivors were counted and measured for SL (nearest 0.01 mm) via calibrated digital images (Image Pro Premier V9.0, Media Cybernetics ® , Rockville, MD, USA).

CO 2 and Temperature Levels
We applied a target CO 2 level of 400 µatm (~8.15 pH) for control treatments, a level characteristic of the open ocean and of coastal systems at the onset of the silverside spawning season (spawning typically begins early April and extends through July) (Figure 1 [61]).The target level for high CO 2 was 2200 µatm (~7.50 pH), a level that is commonly experienced by silverside offspring in late spring and summer (Figure 1), but also represents the maximum prediction of average OA for the next 300 years [67] and therefore a common benchmark in many OA studies [61][62][63].The target level for the extreme CO 2 treatment was 6000 µatm (~7.15 pH) during experiment 1, but was reduced to 4200 µatm (~7.20 pH) for experiments 3 and 5.These represent extreme CO 2 conditions rarely reached in contemporary coastal systems, but may become more common under future climate and eutrophication scenarios [33].latitude of our source populations (~41° N), silverside spawning habitats rarely reach temperatures of 28 °C , however, these conditions may become more common given projected increases of 2-3 °C in global mean ocean temperature [68].The optimal culturing temperature for M. menidia from northern latitudes is ~24 °C ; thus, 20 °C and 24 °C treatments were considered near-optimal temperatures, while 17 °C and 28 °C treatments represented sub-optimal thermal conditions [60].We administered four temperature treatments over the course of the five experiments; 17, 20, 24 and 28 • C. The first three temperatures represent local conditions found during the onset (late-April), peak (early-June), and end (July) of the silverside spawning season, respectively (Figure 1).At the latitude of our source populations (~41 • N), silverside spawning habitats rarely reach temperatures of 28 • C, however, these conditions may become more common given projected increases of 2-3 • C in global mean ocean temperature [68].The optimal culturing temperature for M. menidia from northern latitudes is ~24 • C; thus, 20 • C and 24 • C treatments were considered near-optimal temperatures, while 17 • C and 28 • C treatments represented sub-optimal thermal conditions [60].

CO 2 × Temperature Manipulations and Measurements
All experiments followed established best practices and guidelines for seawater acidification in OA research [27].For 2 × 2 and 3 × 2 factorial designs (see Table 1 for overview of experiments and designs), replicate rearing-containers were placed into large temperature-controlled water baths.Elevated CO 2 levels were achieved via gas proportioners (ColeParmer ® , Vernon Hills, IL, USA) mixing air with 100% CO 2 (bone dry grade) that was delivered continuously to the bottom of each replicate rearing-container via airstone.To counteract metabolic CO 2 accumulation, control CO 2 conditions were achieved by forcing compressed laboratory air through a series of CO 2 -stripping units containing granular soda lime (AirGas ® , Waterford, CT, USA), a particle filter (1 µm), and then to each rearing-container via airstone.Target pH levels were monitored daily using a handheld pH probe (Orion Ross Ultra pH/ATC Triode with Orion Star A121 pH Portable Meter (Thermo Fisher Scientific ® , Waltham, MA, USA); Intellical PHC281 pH Electrode with HQ11D Handheld pH/ORP Meter (Hach ® , Loveland, CO, USA) calibrated bi-weekly with National Institute of Standards and Technology (NIST) certified 2-point pH references.Continuous bubbling maintained dissolved oxygen (DO) saturation (>8 mg/L) in rearing vessels.Target treatment temperatures were controlled by thermostats (Aqualogic ® , San Diego, CA, USA) which powered chillers (DeltaStar ® , Lynchburg, VA, USA) or glass submersible heaters to maintain water bath temperatures.
For 3 × 3 factorial experiments, we constructed an automated acidification system composed of nine discrete recirculation-units designed for larval fish rearing.Each recirculating-unit consists of a sump (90 L), a header tank (40 L) and a main tank (240 L) that holds up to five replicate rearing-containers (20 L) fitted with screened overflow holes (100 µm).In these units, seawater continuously circulates from the sump through a UV sterilizer into the header tank, where it is gravity fed to the bottom of each rearing-container, from which it overflows in the main tank and back into the sump.We designed a LabView (National Instruments ® , Austin, TX, USA) based program to fully automate the control of seawater chemistry.The software interfaces with the recirculating-units via a data-acquisition module (NI cDAQ-9184, National Instruments ® ), which controls nine sampling-pumps (one per tank) and a series of gas and water solenoid valves, while receiving input from a central pH electrode (Hach pHD ® digital electrode calibrated weekly using NIST 2-point pH references) and DO probe (Hach LDO ® Model 2).The software sequentially assesses the pH conditions in each recirculating-unit (once per hour) by pumping water for ~7.5 min through the housing of the central pH probe, comparing measured pH levels to set-points and then adjusting levels by bubbling standardized amounts of 100% CO 2 (bone dry grade, AirGas ® ) or CO 2 -stripped air into the sump of each tank.The software also maintains DO saturation (>8 mg/L) by bubbling in CO 2 -stripped air.LabView logs current pH, temperature, and DO conditions before cycling to the next unit.Temperatures were controlled by thermostats (Aqualogic ® ) that powered submersible heaters or in-line chillers (DeltaStar ® ).
Actual treatment CO 2 levels were determined based on measurements of pH, temperature, salinity, and total alkalinity (A T ).Treatment tanks were sampled three times per experiment for measurements of A T (µmol kg −1 ).Seawater was siphoned and filtered (to 10 µm) into 300-mL borosilicate bottles.Salinity was measured at the time of sampling using a refractometer.Bottles were stored at 3 • C and measured for A T within two weeks of sampling using an endpoint titration (G20 Potentiometric Titrator, Mettler Toledo ® , Columbus, OH, USA).Methodological accuracy (within ±1%) of alkalinity titrations were verified and calibrated using Dr. Andrew Dickson's (University of California San Diego, Scripps Institution of Oceanography, https://www.nodc.noaa.gov/ocads/oceans/Dickson_CRM/batches.html) certified reference material for A T in seawater.The partial pressure and fugacity of CO 2 (pCO 2 , f CO 2 ; µatm) as well as dissolved inorganic carbon (C T ; µmol kg −1 ) and carbonate ion concentration (CO 3 2− ; µmol kg −1 ) were calculated in CO2SYS (V2.1, http://cdiac.ornl.gov/ftp/co2sys)based on measured A T , pH, temperature, and salinity using K1 and K2 constants from Mehrbach et al. [69] refitted by Dickson and Millero [70] and Dickson [71] for KHSO 4 .An overview of pH and carbonate chemistry measurements for each experiment is given in Table 2.

Response Traits and Statistical Analysis
For all replicates in each experiment we quantified time (d) to first-hatch (day of fertilization to day of first-hatch), the % of embryo survival (fertilization to hatch), the % of larval survival (hatch to end of experiment), SL at hatch, and post-hatch growth rate ((mean final SL − mean hatch SL)/number days reared post-hatch).For experiment 1, only survival traits were quantified.Time to first-hatch was invariant between CO 2 levels and thus was not analyzed statistically.Proportional survival data were logit transformed [=log 10 (survival/(1 − survival))] prior to analysis [72].Grubb's test [73] was used to identify potential outlying replicates, resulting in the removal of three replicates throughout the dataset for low embryo survival (Grubb's test, p < 0.05).
Statistical analyses were computed using SPSS (V20, IBM).As a first step, we used linear mixed effects models incorporating data from all experiments to test for significant effects (α < 0.05) of CO 2 , temperature, their interaction (fixed factors) and experiment (random factor) for each response trait: Response trait = CO 2 + temperature + CO 2 × temperature + experiment + error.
Response trait data were checked for variance homogeneity and assumption of normality using Levene's and Shapiro-Wilk tests (α < 0.05), respectively.If linear mixed effects models identified traits with significant CO 2 or CO 2 × temperature effects, we used two-way analysis of variance (ANOVA) to test for significant effects of CO 2 , temperature, and their interaction within each experiment: Response trait = CO 2 + temperature + CO 2 × temperature + error.This approach was implemented to characterize how CO 2 effects differed between experiments.If two-way ANOVAs detected significant (α < 0.05) CO 2 or CO 2 × temperature interactive effects, we used one-way ANOVAs to test for significant CO 2 effects within temperature treatments.Where necessary, least-significant-difference (LSD) post-hoc tests were used for multiple comparisons.We conducted two-and one-way ANOVAs on experiment 1 separately because the extreme CO 2 level implemented there (~6000 µatm) was higher than in experiments 3 and 5 (~4200 µatm).ANOVA groups were checked for variance homogeneity and assumption of normality using Levene's and Shapiro-Wilk tests (α < 0.05), respectively.
Linear mixed effects models indicated that response traits were highly variable between experiments.Hence, we implemented an additional approach to better describe CO 2 effects across temperature treatments.We quantified the temperature-specific CO 2 effect sizes for each trait (T) for each experiment by calculating the log-transformed CO 2 response ratio (lnRR) to high and extreme levels of CO 2 exposure.LnRRs evaluate the average proportional change in a trait relative to control treatments, with negative lnRRs indicating negative CO 2 effects.LnRRs have become a common metric for evaluating CO 2 effects in meta-analyses when comparing variable responses across studies [6,25,74].LnRRs were calculated as: lnRR(T) = ln(T high or extreme CO2 ) − ln(T control CO2 ).
Overall temperature-specific CO 2 responses were calculated as the mean lnRR(T) across experiments.

Overall CO 2 Effect Size (LnRR)
The overall CO 2 effect on embryo survival was small in response to high CO 2 conditions (within ±0.06) and similar across temperature treatments (Figure 3A).For offspring exposed to extreme CO 2 , all responses were negative (−0.04 to −0.30), but there was no apparent trend with temperature (Figure 3B).For hatch size, overall effects were small both at high and extreme CO 2 treatments (±0.03), again with no apparent temperature dependency (Figure 3C,D).The overall effect of high CO 2 conditions on larval survival was highly variable (−0.64 to 0.55) and overall neutral across temperatures (Figure 3E).Interestingly, the effect of extreme CO 2 conditions on larval survival was positive at 17 • C (0.42) but became increasingly negative with increasing temperatures (−0.18 at 24 • C and −0.44 at 28 • C, Figure 3F).Average CO 2 effects for growth rate were small (within ± 0.10), but exhibited a dome-shaped response across temperatures at both CO 2 levels, with negative growth responses at sub-optimal rearing temperatures (Figure 3G,H).

Discussion
We conducted five factorial experiments to evaluate the sensitivity of M. menidia early life traits to high (2000-2800 µatm) and extreme CO 2 conditions (4000-6200 µatm) across four temperatures (17,20,24, and 28 • C) that encompassed contemporary and potential future conditions in nearshore silverside spawning habitats.The experiments showed few significant CO 2 effects on response traits.Significant reductions in embryo survival occurred at 17 and 24 • C in a single experiment and at the most extreme CO 2 treatment (~6000 µatm).Effects on hatch length showed evidence for CO 2 × temperature interactions, given that elevated CO 2 reduced hatch length at 24 • C during one experiment, while increasing hatch length at 28 • C during another.There were no significant effects of CO 2 on larval survival or growth rate.Together, these findings suggest that M. menidia offspring can tolerate high to extreme CO 2 levels across most of the species' thermal range.
The apparent CO 2 resilience of M. menidia offspring may reflect the pH/CO 2 variability typical of their nursery habitat.Atlantic silversides spawn in shallow subtropical to temperate estuaries [75] where seasonal acidfication elicits increasingly large diel pH fluctuations while progressively reducing daily mean and minimum pH levels [35].Such patterns of seasonal pH/CO 2 variation appear to be common in shallow nearshore habitats [36].As a batch-spawning fish, silversides spawn fortnightly from late April to early July (at ~41 • N) which coincides with the period of most rapid habitat acidification [61] Thus, a single female will deposit subsequent batches of embryos into a progressively more pH variable and acidic environment.In a previous study, we found offspring CO 2 tolerance closely tracked temporal trends in habitat acidification [61].Transgenerational plasticity is a possible explanation for this rapid shift, by which adults experiencing a progressively more acidic environment augment offspring phenotypes to better match current environmental conditions [76].An additional source of CO 2 tolerance may arise from local adaptation.Despite being an annual fish with high population connectivity, Atlantic silversides exhibit local adaptation for traits involved in growth and environmental sex determination [77], which are likely maintained through the continuous selection of locally suited genotypes [78].As previously demonstrated, early life survival under high CO 2 conditions is a heritable trait in this species, suggesting that CO 2 tolerance could evolve [63].Local adaptation to acidified habitats through the selection and maintenance of CO 2 -tolerant traits has been demonstrated in other taxa [29].For Atlantic silversides that spawn in habitats prone to acidification, adaptations that enable high-CO 2 tolerance are likely well represented in wild populations and would explain the observed CO 2 tolerance.Importantly, we found that exposure to ~6000 µatm pCO 2 did reduce embryo survival during experiment 1, while offspring were largely tolerant to ~4200 µatm in subsequent experiments.While 6000 µatm is an extreme, likely unrealistic CO 2 level for silverside spawning habitats, it may represent a tolerance threshold for M. menidia.Identifying such thresholds are necessary to accurately assign an organisms' sensitivity to future climate change [79].
Maternal provisioning of eggs through modifications of energy content or fatty acid composition may further influence offspring CO 2 sensitivity [58].Such differences may have contributed to CO 2 effects on hatch length documented during experiments 3 and 5.In Atlantic cod, CO 2 induced reductions in hatch size were not accompanied by increased utilization of yolk reserves during embryogenesis [80].The authors suggest that yolk utilization was already maximized, and increased demands on acid/base regulation resulted in a shift of endogenous energy use away from somatic growth.Conceivably, CO 2 reductions in hatch size during experiment 3 were the result of a similar mechanism.Conversely, differences in maternal provisioning of embryos from experiment 5 may have stimulated yolk utilization under elevated CO 2 , leading to increased embryonic growth and hatch size [13].Fish embryos passively experience their environment with fixed energy reserves [81] but are likely most sensitive to elevated CO 2 [11].Further investigations are needed into how CO 2 × temperature combinations influence embryo energetics.
The apparent tolerance of M. menidia offspring to combined climate stressors contrasts with the growing evidence for compounding effects of near future OA and warming in the early life stages of other fish species.For example, combined treatments synergistically decreased embryo survival in Antarctic dragon fish (Gymnodraco acuticeps) [48] and compromised temperature acclimation and aerobic performance in emerald rockcod (Trematomus bemacchii) [49].As extreme stenotherms, polar species appear particularly vulnerable to combined climate effects [82], but eurythermal temperate species have demonstrated similar sensitivities.For example, exposure to acidification and warming reduced hatch size and larval survival in the Senegalese sole (Solea senegalensis) [14] and Atlantic cod (Gadus morhua) [80].In the congeneric M. beryllina, a large reduction in survival was found when simultaneously exposed to high-CO 2 and 29 • C [59].The CO 2 × temperature tolerance demonstrated by M. menidia offspring is likely a manifestation of conditions widely experienced by wild silverside early life stages.The acidification of their near shore nursery habitat is largely driven by seasonal changes in community respiration that generally peak with seasonally maximum water temperatures [36].Thus, simultaneous occurrence of potentially stressful temperature and CO 2 levels are a regular feature of M. menidia spawning habitat.Furthermore, because seasonal habitat changes are of the same direction and similar magnitude to climate projections, existing phenotypic or genetic variation already present in silverside populations may confer some degree of tolerance to future marine climate change [83].
While M. menidia early life stages appear resistant to elevated CO 2 across a broad thermal regime, the addition of other stressors could potentially be detrimental.For example, temperature-dependent metabolic processes that drive coastal acidification simultaneously consume oxygen; hence, warming, acidification, and hypoxia co-occur in M. menidia nursery habitats [36,84].Given their co-occurrence in nature, physiological responses to elevated CO 2 and low DO are likely connected.Intermediate CO 2 exposure can elicit important adaptive responses which may mediate sublethal effects of low DO [84], yet more extreme exposures may act synergistically to elevate stressor sensitivity [85].Thus, factorial CO 2 × DO × temperature experiments would be insightful for more robust characterizations of coastal climate effects on fish early life stages.
Whole lifecycle effects of elevated CO 2 exposure remain critically understudied in fish [25].While acclimation to chronic hypercapnia likely has small metabolic costs [86], over longer timescales tradeoffs associated with increased acid/base regulation could compromise other physiological processes [43].In a previous study, we documented small but significant size reductions in M. menidia reared under ~2200 µatm CO 2 and 17 • C for approximately a third of their lifespan [62].Importantly, differences in length were only detected after two months of continuous high-CO 2 exposure.In the present study, CO 2 effect sizes calculated for growth rates displayed dome-shaped response curves, with more negative responses at sub-optimal rearing temperatures.For offspring reared under 17 • C and high CO 2 , the average growth effect size was −0.08 (i.e., −8%), a response of similar magnitude to previously documented growth reductions under the same conditions after four months [62].Importantly, that study used large sample populations (>2000 individuals) providing the necessary power to statistically confirm a CO 2 effect.Arguably, many early-life experiments with smaller sample sizes lack the power to robustly detect small effects [87].Thus, it is possible that small or undetectable CO 2 reductions in early-life growth accrue and become detectable during long-term exposures.Even minor changes to early life development may have important carry-over effects to later life stages and ultimately impact fitness [88].As an annual fish, juvenile growth during summer is critically important for M. menidia, as larger individuals have higher overwintering survival [89].How warming temperatures may interact with CO 2 over longer time-scales is presently unknown and represents a serious gap in our understanding of how combined climate stressors will impact fish [90].
Across experiments, CO 2 responses were highly complex, consistent with previous OA studies on silverside offspring [58,59,61].Experiments produced functionally different outcomes within equivalent treatment conditions despite meticulously controlled experimental conditions.For all traits but growth rate, inter-experiment variation was more substantial than variability driven by CO 2 or temperature level.A portion of this variability could be elicited by small differences in food quantity or quality, water source, or realized CO 2 levels, but parentage likely constitutes the largest source of

Figure 1 .
Figure 1.Average mean (+/−minimum/maximum) monthly temperature and pH conditions during the spawning and growing season of Atlantic silversides in (A) Flax Pond, Long Island, New York and (B) Mumford Cove, Connecticut.The sites provided wild spawners for experiment 1 (A) and experiments 2-5 (B).Long-term averages were derived from monitoring data collected in 15 min intervals by (A) USGS station #01304057 between 2008 and 2018 and (B) the Baumann lab in Mumford Cove between 2015-2018.

Figure 1 .
Figure 1.Average mean (+/−minimum/maximum) monthly temperature and pH conditions during the spawning and growing season of Atlantic silversides in (A) Flax Pond, Long Island, New York and (B) Mumford Cove, Connecticut.The sites provided wild spawners for experiment 1 (A) and experiments 2-5 (B).Long-term averages were derived from monitoring data collected in 15 min intervals by (A) USGS station #01304057 between 2008 and 2018 and (B) the Baumann lab in Mumford Cove between 2015-2018.

Figure 2 .
Figure 2. M. menidia.Offspring responses to control (blue), high (red), and extreme (green) CO2 conditions at four temperatures across five CO2 × temperature factorial experiments.Traits include embryo survival (A-E), hatch length (F-I), larval survival (J-N) and larval growth rate (O-R).Individual replicates are represented by small faded circles.Treatment means (±SD) are depicted by large, bold circles and connected by dotted lines.Note: different scales used for hatch length measurements due to differences in sample timing; panels F and G use 1dph length Y axis (left) while panels H and I use hatch length Y axis (right).

Figure 2 .
Figure 2. M. menidia.Offspring responses to control (blue), high (red), and extreme (green) CO 2 conditions at four temperatures across five CO 2 × temperature factorial experiments.Traits include embryo survival (A-E), hatch length (F-I), larval survival (J-N) and larval growth rate (O-R).Individual replicates are represented by small faded circles.Treatment means (±SD) are depicted by large, bold circles and connected by dotted lines.Note: different scales used for hatch length measurements due to differences in sample timing; panels F and G use 1dph length Y axis (left) while panels H and I use hatch length Y axis (right).

Figure 3 .
Figure 3. M. menidia.CO2 effect sizes using log-transformed response ratios (lnRR) of high (light grey) and extreme (dark grey) CO2 exposure across four rearing temperatures.Response traits include embryo survival (A,B), hatch length (C,D), larval survival (E,F), and growth rate (G,H).Circles represent lnRRs of each experiment, while black lines represent lnRRs averaged across experiments at each rearing temperature.Negative (positive) values indicate a trait decrease (increase) at elevated CO2 levels compared to control CO2 conditions.

Figure 3 .
Figure 3. M. menidia.CO 2 effect sizes using log-transformed response ratios (lnRR) of high (light grey) and extreme (dark grey) CO 2 exposure across four rearing temperatures.Response traits include embryo survival (A,B), hatch length (C,D), larval survival (E,F), and growth rate (G,H).Circles represent lnRRs of each experiment, while black lines represent lnRRs averaged across experiments at each rearing temperature.Negative (positive) values indicate a trait decrease (increase) at elevated CO 2 levels compared to control CO 2 conditions.
• 56.85 N, 73 • 6.15 W), and the experiment took place at Stony Brook University's Flax Pond Marine Laboratory.During 2016 and 2017 (experiments 2-5), spawning adults were collected from Mumford Cove (41 • 19 25 N 72 • 01 07 W), and experiments were conducted in the Rankin Seawater Facility at University of Connecticut's Avery Point campus.Ripe adults were held overnight at 20 • C in aerated tanks at low densities with no food and strip-spawned the next day.Fertilization dates for each experiment are reported in Table .

Table 1 .
Summary of five CO 2 × temperature experiments on M. menidia offspring.Treatment levels for pCO 2 (µatm) and temperature ( • C) represent target conditions, actual measured values are presented in Table2.Trait are abbreviated as embryo survival (ES), hatch length (HL), larval survival (LS), and growth rate (GR).

Table 4 .
Summary statistics for linear mixed models testing the effects of CO2, temperature and their interaction (fixed factors) and experiment (random factor) on four response traits; embryo survival (ES), hatch length (HL), larval survival (LS), and growth rate (GR) of M. menidia offspring.Significant (α < 0.05) factors are denoted by p-values in bold.

Table 4 .
Summary statistics for linear mixed models testing the effects of CO 2 , temperature and their interaction (fixed factors) and experiment (random factor) on four response traits; embryo survival (ES), hatch length (HL), larval survival (LS), and growth rate (GR) of M. menidia offspring.Significant (α < 0.05) factors are denoted by p-values in bold.