Oxidative stress effects are not correlated with differences in heat tolerance among congeners of Mytilus

Elucidating the mechanisms that establish thermal optima and limits can shed light on the factors governing organisms' distribution patterns and their sensitivities to global change (Dong et al., 2022; Pörtner, 2002; Somero et al., 2017; Williams et al., 2016). Adaptive evolved differences in thermal sensitivities of the major classes of macromolecules – proteins, nucleic acids and membranes – play a central role in thermal adaptation (Dong and Somero, 2009; Somero et al., 2017), but it remains less clear to what extent other mechanisms might also contribute to adaptive differences among species. Oxidative stress figures prominently among the other mechanistic factors that have been hypothesized to contribute to species differences in thermal sensitivity (Abele et al., 1998; Feidantsis et al., 2020; Pörtner, 2010; Tomanek, 2011), although the available evidence does not uniformly support a direct link between oxidative stress and thermal tolerance (e.g. Mueller et al., 2014). Oxidative stress results from the imbalance between the production of highly reactive metabolic byproducts (e.g. reactive oxygen species, ROS) and scavenging of these harmful byproducts by antioxidant systems before they inflict macromolecular damage to nucleic acids, lipids or proteins (Halliwell and Gutteridge, 2007). If left unrepaired, oxidative damage can greatly impair cellular function. It is often observed that ROS production and/or oxidative damage markers increase at elevated body temperatures (Abele et al., 1998, 2001; Bagnyukova et al., 2007; Gleason et al., 2017; Heise et al., 2006, 2003; Jimenez et al., 2016; Wang et al., 2018). Because future environmental patterns are expected to include more unpredictable and more extreme temperatures (e.g. Harris et al., 2018; Stillman, 2019), it is imperative to understand more fully the mechanistic underpinnings of thermal tolerance, including the role of oxidative stress.

The rocky intertidal environment has proven to be an excellent system for studying adaptive differences in environmental stress tolerance. The likelihood that oxidative stress also may exert a strong impact on intertidal zone animals stems from patterns of environmental variation experienced by these species. Dictated by the rhythms of the tides and their interactions with solar and atmospheric factors (Denny and Harley, 2006; Helmuth, 1998), this habitat exposes organisms to episodic extremes of temperature, as well as wide variation in oxygen availability and food supply. Variations in these key environmental factors are found across latitude and at much smaller scales, such as across vertical gradients. Not surprisingly, even closely related inhabitants of this dynamic habitat display sometimes striking differences in their susceptibility to abiotic stressors including temperature (e.g. Dong et al., 2022, 2008; Dowd and Somero, 2013; Stillman and Somero, 2000; Tomanek and Helmuth, 2002; Tomanek and Somero, 2000).

The three congeners of Mytilus mussels found on the Pacific coast of North America vary dramatically in several aspects of thermal physiology, from the molecular to the organismal level (reviewed in Lockwood and Somero, 2011), which makes them a particularly useful system for studying the mechanistic basis of inter-specific differences in thermal tolerance. Mytilus trossulus Gould is less heat tolerant than Mytilus galloprovincialis Lamarck or Mytilus californianus Conrad, as illustrated by differences in survival after acute, high-temperature episodes (Dowd and Somero, 2013) or following chronic exposure to moderate temperatures (Schneider, 2008). The heat-tolerant M. galloprovincialis has a lower resting heart rate and a higher critical temperature for cessation of cardiac activity than M. trossulus (Braby and Somero, 2006b); the critical temperature for cessation of cardiac activity is even higher for M. californianus (Logan et al., 2012). At the level of single macromolecules, a handful of metabolic enzymes of M. galloprovincialis, as well as cytosolic malate dehydrogenase for M. californianus, have been shown to maintain the appropriate configuration for substrate binding affinity at higher temperatures, unlike the same enzymes from M. trossulus (Fields et al., 2006; Lockwood and Somero, 2012). The species also vary in their molecular responses to heat stress, such as the threshold for induction of synthesis of heat-shock proteins and patterns of protein ubiquitination (Buckley et al., 2001; Hofmann and Somero, 1996). Comparative transcriptomic and proteomic analyses of M. trossulus and M. galloprovincialis have further characterized the temperature-dependent expression profiles that might underpin their differences in heat tolerance. Specifically, these two species differ in their relative induction of small heat shock proteins that stabilize the cytoskeleton and, most relevant to this study, in their induction of constituents of the antioxidant system (Fields et al., 2012; Lockwood et al., 2010; Tomanek and Zuzow, 2010). Genomic sequences for chaperones and antioxidant system proteins also appear to be under strong selection in the more heat-tolerant M. galloprovincialis relative to M. trossulus (Popovic and Riginos, 2020). These specific examples of differences in expression patterns and biochemistry are reflected in a higher frequency of stabilizing amino acid substitutions across the proteome of M. galloprovincialis than in M. trossulus (Popovic and Riginos, 2020).

The evolutionary and biogeographic history of the genus Mytilus has also been reconstructed (Koehn, 1991; Seed, 1992), providing important context for the interpretation of inter-specific physiological differences (Lockwood and Somero, 2011). The California mussel, M. californianus, arose in the relatively cool waters of the eastern Pacific. In this region, M. californianus experiences episodic exposure to high body temperatures only at low tide (Denny et al., 2011). The blue mussel M. trossulus also originated in the eastern Pacific. Finally, the blue mussel M. galloprovincialis diverged more recently from the Mytilus edulis lineage in the more saline and warmer Mediterranean Sea (Seed, 1992). These evolutionary events have been followed by a period of human-mediated spread, through both intentional (e.g. introductions for aquaculture) and unintentional means (e.g. transport in ballast water; Carlton and Geller, 1993; Wilkins et al., 2008). Mytilus galloprovincialis has been a particularly successful invader around the globe (Branch and Steffani, 2004; Daguin and Borsa, 1999), and in the eastern Pacific it has displaced native M. trossulus throughout the southern, warmer portion of its former range (McDonald and Koehn, 1988). Enhanced heat tolerance of M. galloprovincialis is a key factor in its invasive success (Lockwood and Somero, 2011).

Whereas differences in thermal responses at the level of cardiac function, gene expression and protein thermal stability have been shown to correlate with whole-organism heat tolerance and with global (re)distribution patterns of Mytilus congeners, the putative contribution of oxidative stress resistance to these inter-specific patterns has not been functionally evaluated. The factors influencing oxidative stress in many intertidal animals are complex, however, because episodes of elevated body temperature tend to coincide with times when oxygen availability is low. For example, in intertidal mussels, high body temperatures occur when the tide is low, air temperature and/or solar irradiance is high, and wave action is minimal. To counter the detrimental consequences of desiccation, many intertidal animals close the shell valves (Miller and Dowd, 2017) or otherwise seal up during aerial episodes of heat stress. This behavioral response limits the availability of oxygen to the tissues and forces reliance on less efficient anaerobic pathways in the cytosol (Bayne et al., 1976; Shick et al., 1988; Widdows and Shick, 1985). Hypoxia is known to coincide with ROS production in a variety of aquatic organisms (Welker et al., 2013), and the combination of hypoxia and acute temperature increases might be expected to inflate levels of oxidative stress (Castro et al., 2020; Sun et al., 2020). For example, M. galloprovincialis gill cells produced more ROS upon exposure to 30°C in air than to 15°C in air (Wang et al., 2018). Other toxic byproducts of glycolysis such as methylglyoxal (derived from glucose) might also be produced. In addition, mussels and other intertidal organisms quickly reopen the shell valves upon reimmersion (Dowd and Somero, 2013; Miller and Dowd, 2017). This behavior coincides with a drop in body temperature and leads to a rapid influx of oxygen that is analogous to reperfusion after ischemia. This rapid reperfusion has been identified as a key contributor to oxidative macromolecular damage in stroke and disease models (Li and Jackson, 2002), as well as in comparative physiology models (Bickler and Buck, 2007; Lushchak et al., 2005), and it could represent a key period of oxidative stress in mussels (Rivera-Ingraham et al., 2013). For example, Gracey et al. (2008) identified a temporal pattern of gene expression in M. californianus that may be linked to cycles of oxidative stress. Furthermore, in behavioral studies, we have observed that all three species studied here close up during acute heat stress, but the more heat-sensitive congener M. trossulus appears to avoid further cycles of hypoxia–reoxygenation by remaining open during recovery periods between stressful temperature episodes. This behavioral evidence, combined with the omics results described above, led us to hypothesize that M. trossulus suffers more oxidative damage than its more heat-tolerant congeners during episodes of heat stress (Dowd and Somero, 2013).

Building on this substantial background knowledge, here we addressed two outstanding questions regarding inter-specific differences in acute heat tolerance among Mytilus congeners. First, what role do oxidative stress and oxidative damage to macromolecules play in these observed differences? To more directly address the hypothesis that Mytilus congeners vary in resistance to oxidative damage during/after exposure to episodes of heat stress, we compared several functional measures across the three Mytilus species: activities of two antioxidant defense enzymes (catalase and superoxide dismutase); and the amount of lipid oxidation and DNA oxidation, and two-dimensional gel-based proteomics measures of oxidative protein damage (amino acid side-chain carbonylation). We predicted that more heat-tolerant congeners would have higher antioxidant capacities and/or accumulate less oxidative damage from acute heat stress episodes. We also expected that the heat-sensitive M. trossulus might accumulate oxidative damage to key proteins that would otherwise play protective roles during/after heat stress; carbonylation is known to impede protein function (e.g. Grimsrud et al., 2007; Pierce et al., 2008). Second, how do metrics of oxidative stress in these mussels change as a result of repeated exposure to thermal stress in air versus seawater? Earlier Mytilus omics studies examined responses to single episodes of heat stress in seawater only (Lockwood et al., 2010; Tomanek and Zuzow, 2010); we extended our analyses to include one and three consecutive daily episodes of thermal stress in seawater or in air. The experimental design employed here better reflects conditions as they are experienced by mussels in the field (Denny et al., 2011; Miller and Dowd, 2017), and it attempts to capture species differences that might not appear after a single episode or when mussels are immersed in seawater throughout the exposure. Recent monitoring and modeling studies show that thermal stress events often cluster into multi-day windows for intertidal species; these relatively rare ‘heat waves’ coincide with midday low tides when air temperature is high and wave action is low (Harley, 2008; Mislan and Wethey, 2015; Seuront et al., 2019). Moreover, both behavior and survival of these Mytilus congeners exhibit clear differences between thermal stress in air and thermal stress in seawater (Dowd and Somero, 2013), and ecological and genetic studies suggest that the outcomes of competition among Mytilus congeners can vary between these environmental contexts (McDonald et al., 1991; Schneider and Helmuth, 2007; Wonham, 2004).

The results reported here highlight numerous differences in the physiological consequences of heat stress in air versus heat stress in seawater; these differences may help to explain idiosyncrasies of Mytilus distribution and invasion patterns. However, the results largely contradict predictions of the hypothesis that oxidative stress mediates acute heat tolerance, calling into question its relevance in this study system.

We studied three Mytilus congeners from the Pacific coast of North America that vary in their heat tolerance. Mytilus trossulus (heat sensitive) and M. galloprovincialis (heat tolerant) have been the subject of extensive comparative studies outlined above; we included heat-tolerant M. californianus to allow stronger inferences and to avoid spurious patterns in a two-species comparison (Garland and Adolph, 1994). Adult M. californianus were collected from a single location in the middle rocky intertidal zone at Hopkins Marine Station (HMS) in Pacific Grove, CA, USA (36.6216°N, 121.9042°W). In order to avoid hybrid blue mussels – HMS sits within the hybrid zone for the native and invasive congeners (Braby and Somero, 2006a; Hilbish et al., 2010) – M. trossulus and M. galloprovincialis were collected from docks outside the hybrid zone in Newport Harbor, OR, USA (44.6226°N, 124.0520°W), and Santa Barbara Harbor, CA, USA (34.4089°N, 119.6893°W), respectively, and transported to HMS over ice and in air-filled coolers. Based on the collection site features, the blue mussels had been acclimatized in the wild to constant immersion, whereas M. californianus had been acclimatized to rhythmic immersion and emersion according to the lunar cycle at HMS. All three species were subsequently held under common conditions of constant immersion in flow-through seawater tanks at HMS (salinity 34±1 ppt; temperature 12–15°C) for at least 4 weeks – and no longer than 12 weeks – prior to the start of any experiment. This allowed sufficient time for acclimation to common-garden conditions (Buckley et al., 2001). Although field studies have documented differences in mean size among these species where they co-occur (Braby and Somero, 2006a; Dutton and Hofmann, 2008), experiments herein were carried out on individuals of comparable size (M. californianus 61–89 mm; M. galloprovincialis 55–78 mm; M. trossulus 50–78 mm). Sex of the animals was not determined. Mussels were fed 3 times a week with a mixture of coccolithophorids (Isochrysis spp., Pavlova spp.), green algae (Tetraselmis spp.) and diatoms (Thalassiosira weissflogii) (Shellfish Diet 1800^©, Reed Mariculture). Mussels were not fed during the 3 days of the episodic heat stress exposures.

Because the two blue mussels cannot be distinguished morphologically, we verified the putative species identifications for those species (M. trossulus from Newport, OR, USA; M. galloprovincialis from Santa Barbara, CA, USA) using the polymorphic molecular markers Glu-5′ and ITS-1. These marker regions were amplified with established primers (Heath et al., 1995; Rawson et al., 1996). Only one putative M. trossulus was mis-identified; its molecular markers were consistent with M. galloprovincialis. This individual was excluded from analyses.

Heat ramp experiments were conducted in the same manner for each species in both seawater and air (as in Dowd and Somero, 2013). Seawater heat ramps were conducted in an insulated cooler. Seawater temperature was controlled with a programmable laboratory water bath temperature controller (Cole-Parmer Polystat^®). The outflow freshwater circulation from the water bath was passed through a countercurrent heat exchanger, while a pump in the cooler passed seawater in the opposite direction through the heat exchanger and back to the cooler. Air heat ramps were carried out in insulated polystyrene boxes in a humidified, walk-in chamber using established methods (Logan et al., 2012). Air temperature was regulated with a programmable benchtop temperature controller (Omega CSC32) connected to a mat heating element (Omega Model SRMU020230, 120 V, 1.25 A, 5.08×76.2 cm). A fan inside the box ensured adequate air mixing, and ice packs in the boxes were used to dampen oscillations in the air temperature otherwise caused by cycling of the heating element. In all cases, the animals were held at a control temperature (13–14°C) within the acclimation range until just before the start of the heat ramp. Temperatures were increased and decreased gradually from 13–14°C at 16°C h⁻¹ for all species; heating rates as high as 20.2°C h⁻¹ have been recorded for intertidal M. californianus at HMS on the warmest days, but they are more typically less than 8°C h⁻¹ (Denny et al., 2011; Miller and Dowd, 2017). In each experiment, the animals were held at the peak temperature (seawater 33±1°C; air 32±1°C) for 45 min before cooling began. The peak temperature of 32–33°C approximates the highest temperature measured at a low-shore site at HMS (Miller and Dowd, 2017). Mussels were returned to control temperature seawater immediately at the conclusion of the heat ramp. In each experiment, individuals were sampled after exposure to one or three heat ramps. Heat waves of 3 days are typical of the shoreline at HMS (Denny et al., 2011). Successive ramps started at the same time (between 12:00 h and 13:00 h) on subsequent days. For simplicity, we refer to treatments as SWX and AirX, where X is the number of heat stress episodes (1 or 3).

Individuals in each treatment group were sampled immediately prior to the beginning of the next day's heat ramp. Thus, they were given ∼20 h of recovery in control conditions. There were two reasons for this design. First, if oxidative damage does contribute to species differences in survival, which play out over days to weeks (Dowd and Somero, 2013), we would expect metrics of damage to accumulate and persist in the more heat-sensitive congener. Second, the design captures mussels' physiological states at the time when the next stressful period of heat/immersion would begin.

We focused our analyses on gill tissue. Gill has high ciliary activity for suspension feeding (Jørgensen, 1974) and is the primary site of oxygen uptake, so we expected that consequences of changes in oxygen availability might be most pronounced in this tissue. Furthermore, gill function is essential for nitrogenous waste excretion (Thomsen et al., 2016). Thus, damage to gill tissue could have pronounced effects on the whole organism. Gill has also been the focus of earlier Mytilus transcriptomics and proteomics studies (Gracey et al., 2008; Lockwood et al., 2010; Tomanek and Zuzow, 2010), providing important background for comparison with the present results.

At the appropriate time in each experiment, gill tissue of each individual was dissected out and flash-frozen in liquid nitrogen, after which tissues were stored at −80°C until analysis. Aside from the DNA oxidation assay, all subsequent assays maintained the individual mussel as the unit of analysis (i.e. samples were not pooled).

All enzymatic assays were conducted at a standard cuvette temperature of 25±0.2°C on a UV-1601 spectrophotometer (Shimadzu, Kyoto, Japan), equipped with a temperature-controlled cell that was regulated by a Lauda RM6 recirculating water bath (Brinkmann, Westbury, NY, USA). Tissues were homogenized in a bead homogenizer at a constant 1:4 ratio of tissue to potassium phosphate buffer (50 mmol l⁻¹, pH 6.8 at 4°C), then centrifuged at 18,000 g and 4°C for 30 min. The supernatant was removed, centrifuged a second time for 5 min, and then separated into aliquots for the different enzyme activity assays.

Total superoxide dismutase (SOD; EC 1.15.1.1) activity (including both the cytosolic Cu-Zn and mitochondrial Mn-dependent forms) and catalase (CAT; EC 1.11.1.6) activity were assayed on the same gill homogenates. SOD was assayed in triplicate for each sample using the xanthine oxidase/cytochrome c method (modified from Dowd et al., 2013; McCord and Fridovich, 1969). The reaction mixtures contained 10 µmol l⁻¹ cytochrome c and 50 µmol l⁻¹ xanthine in 50 mmol l⁻¹ potassium phosphate buffer containing 0.1 mmol l⁻¹ EDTA, at pH 7.6. We included 16 U CAT enzyme in each reaction to prevent reoxidation of cytochrome c by the H₂O₂ produced by the SOD reaction (Crapo et al., 1978). Baseline reactions were started by adding xanthine oxidase to generate sufficient superoxide ion (the substrate of SOD) to cause an increase in A₅₅₀ of ∼0.020 min⁻¹ due to the reduction of cytochrome c. One unit of SOD activity was defined as the quantity needed to achieve 50% competitive inhibition of this baseline rate, and activity in mussel samples was standardized per gram fresh mass. CAT activity was measured in triplicate for each sample by following the rate of disappearance of hydrogen peroxide (decrease in absorbance at 240 nm) in the same phosphate buffer solution (Beers and Sizer, 1952; Dowd et al., 2013). One unit of CAT activity was defined as the quantity decomposing 1 µmol of H₂O₂ (molecular extinction coefficient 0.0394 l mmol⁻¹ cm⁻¹) in 1 min at pH 7.6 (note that this is not the standard pH of 7.0 in most of the literature). CAT specific activity was standardized in units of per gram fresh mass.

Lipid peroxidation represents one of the most common forms of macromolecular damage under exposure to oxidative stress. The thiobarbituric acid reactive substances (TBARS) assay was run following established methods (Hermes-Lima and Storey, 1995). Lipids were extracted in 1.1% phosphoric acid and then incubated with 1% thiobarbituric acid (TBA) in 50 mmol l⁻¹ NaOH. Blanks were treated with 3 mmol l⁻¹ hydrochloric acid instead of TBA. Butylated hydroxytoluene (BHT) was included in the assay buffer at 10 µmol l⁻¹ because of its chain-breaking capacity (Pikul et al., 1983); this modification reduces artefactual increases in peroxides induced by heating samples during the preparation. After heating for 15 min at 100°C, lipids were extracted in isobutanol. We measured blank (replacing TBA with 3 mmol l⁻¹ HCl) and TBA-treated absorbance at 532 nm and 600 nm on a microplate reader in duplicate for each sample. Sample equivalents of malondialdehyde (MDA) concentrations were determined from a standard curve of 1,1,3,3-tetramethoxypropane treated in the same fashion; this chemical is converted to the detectable standard MDA via acid hydrolysis during the reaction.

Oxidative damage to DNA was quantified using the New 8-OHdG Check kit (Cosmobio USA), which measures the amount of 8-hydroxy-2′-deoxyguanosine (8-OHdG) in tissues based on competition with a known standard. 8-OHdG is a commonly measured and relatively stable product of DNA oxidation (Baek et al., 2000; Halliwell and Gutteridge, 2007; Liu et al., 2004). Briefly, DNA was extracted from tissues, pooled and quantified into equal total amounts (33 µg) from 5 individuals in each treatment, digested to single nucleotides with P1 nuclease, eluted in buffer and then assayed in duplicate on a microplate with bound anti-8-OHdG antibody, all as per the kit manufacturer's instructions. Concentrations were estimated from a logit standard curve using kit standards. Because of limited nucleic acid material, there was only one replicate per treatment/species combination.

Data for the preceding assays were analyzed in R (http://www.R-project.org/) using linear models that included species, treatment and a species-by-treatment interaction term. For the enzymes, we initially included shell length as a covariate, but likelihood ratio tests showed that this term did not significantly improve the model (SOD P=0.599; CAT P=0.389; TBARS P=0.319). Therefore, shell length was removed from final analyses. SOD activity data were log₁₀ transformed to meet assumptions.

We first attempted to quantify total oxidative protein damage using immunochemical dot-blots for protein carbonylation, a stable marker of protein oxidation (Stadtman, 1993), with an antibody reactive with dinitrophenylhydrazine (DNPH)-derivatized amino acid side-chains (Dalle-Donne et al., 2003). However, we encountered substantial background signal. The DNPH derivatization reaction targets all carbonyl groups regardless of their molecular source (DNA, protein, etc.). We could not abolish this background by treatment of the samples with streptomycin sulfate to precipitate DNA (Luo and Wehr, 2009) or by protein precipitation in 20% trichloroacetic acid (Colombo et al., 2016). Furthermore, these artefacts remained when we excluded the detergents Triton X-100 and Nonidet P-40, which can contain DNPH-reactive impurities and promote the release of contaminating DNA from the nucleus (Luo and Wehr, 2009), from the homogenization buffer. We also removed dithiothreitol (DTT), which via the Fenton reaction with iron can cause artefactual increases in carbonyl measurements (Luo and Wehr, 2009; Stadtman, 1993), with similar lack of effect. We suspect this persistent background results from the accumulation of carbonyl-containing pigments such as fucoxanthins from the algal food. Numerous pigments, many of which contain carbonyl groups, have been identified in Mytilus mussels (Campbell, 1970; Hertzberg et al., 1988), and gill is a heavily pigmented tissue.

We were interested as much in variation in the specific targets of oxidative protein damage among Mytilus congeners as in variation in the total amount of oxidative damage. Therefore, we used a two-dimensional gel proteomics approach with subsequent Western blotting to detect specific carbonylated proteins (Chaudhuri et al., 2006; McDonagh et al., 2005). This technique allowed us to separate and identify individual proteins with stronger (or weaker) carbonyl signal following thermal stress events relative to the control conditions, while excluding background signal from pigments or other impurities. For these analyses, we focused on a comparison of control animals with those subjected to three episodes of heat stress in air or in water (n=4 mussels per treatment per species for proteomics).

The 2D gel electrophoresis protocol followed established methods (Dowd et al., 2010a,b). Briefly, mussel gill proteins were extracted by homogenization in RIPA buffer. Gill protein extracts from each individual were processed in duplicate for all subsequent steps for M. galloprovincialis and M. trossulus; one replicate was used for quantifying protein abundance while the other was used for western blotting for protein carbonylation. For M. californianus, we only processed gels for identifying protein carbonylation. Salts were removed via precipitation in 20% trichloroacetic acid (w/v) in acetone; we processed aliquots of homogenate carrying 750 µg of protein each. Proteins were then resuspended in urea/thiourea rehydration buffer and loaded onto 17 cm immobilized pH gradient (IPG) strips (pH 3–10 non-linear) by overnight passive rehydration at 4°C, prior to isoelectric focusing (IEF) as previously described (Dowd et al., 2010a,b). The two IPG strips for each mussel were run on the same day. In total, two 12-lane IEF runs were needed per species, with treatments evenly distributed across days. IPG strips were frozen at −80°C until further analysis. IPG strip equilibration and reduction (using DTT and IAA), second dimension SDS-PAGE gel runs, and colloidal Coomassie Blue G250 staining were all carried out as previously described for one of the two IPG strips for each individual (Dowd et al., 2010a,b). Prior to equilibration and reduction, the second IPG strip per mussel was washed in 10 mmol l⁻¹ DNPH in 2 mol l⁻¹ HCl for 15 min (McDonagh and Sheehan, 2006); this reaction derivatized protein carbonyl groups to a stable dinitrophenylhydrazone for later detection. The DNPH derivatization reaction was halted by addition of 2 mol l⁻¹ Tris buffer twice for 7 min each. Derivatization at this stage of the process prevents differential IEF of proteins that might be caused by the DNP adduct (Reinheckel et al., 2000), facilitating downstream image processing and comparative analysis of total protein gels with anti-DNPH blots. Proteins were separated in the second dimension on 16 cm (12% acrylamide) SDS-PAGE gels in Bio-Rad Protean II cells at 200 mA constant current for ∼6.5 h. Protein abundance gels were submerged in a colloidal Coomassie Blue G-250 staining solution and gently rotated in the dark overnight, followed by several rounds of destaining in ultrapure water. Gels were scanned on an Epson V750 Professional scanner using the film/transparency adapter at 600 dpi resolution and with automatic gain control turned off.

For each individual mussel, the DNPH-derivatized protein gel was immediately used for large-format western blotting. Proteins were transferred to PVDF membrane overnight at 15 V and 4°C. Membranes were subsequently blocked with 4% blotting grade milk in phosphate-buffered saline containing 0.05% Tween-20 (PBS-T), incubated with primary antibody at 1:10,000 dilution in the blocking solution (rabbit anti-DNP keyhole limpet hemocyanin; Molecular Probes catalog number A-6430), washed 4 times in PBS-T, incubated with secondary antibody at 1:5000 dilution in the blocking solution (goat anti-rabbit HRP-conjugated; Santa Cruz Biotech catalog number sc-2004), and washed again 4 times in PBS-T. Blots were visualized by addition of ECL Reagent (Amersham Pharmacia) for 1 min, followed by exposure to blue light autoradiography film for 10 min (M. galloprovincialis and M. trossulus) or 2 min (M. californianus). Films were immediately scanned for densitometry and image analysis using the same methods as for the gels.

Protein abundance gel and carbonylation blot image analyses were carried out using Delta 2D software v4.0 (Decodon, Germany) as previously described (Dowd et al., 2010a,b). Images for the protein abundance gels and anti-DNP protein carbonylation blots were analyzed separately, because of considerable differences in the number of protein spots visible using the two methods. Images were overlaid manually using the Delta 2D ‘exact’ warping strategy, with extensive manual confirmation of alignment (typically 200–300 manual warping assignments per gel pair). In each type of analysis, spot volumes (intensity×area) were normalized by the total volume of all spots visible in the union image. We replaced the few instances of missing data with a small value of 0.001 to avoid statistical issues introduced by missing values in the data matrix. The normalized data were then arcsine-square root transformed to better meet the assumptions of downstream statistical tests (Dowd, 2012). To enable spot-picking for proteins of interest in the carbonylation analysis from abundance gels, we manually aligned each species' union carbonylation blot image to the union protein abundance image using Delta2D.

The proteomics data were analyzed spot-wise in the Matlab R2017b bioinformatics package using a one-way ANOVA across the three treatments, with a strict alpha value of 0.01 as the threshold statistical significance to avoid false positives. Each species was analyzed separately for each type of data (protein abundance and carbonylation signal). Post hoc comparisons were computed with a least significant difference (LSD) test. As a complementary approach to account for the false discovery rate when conducting numerous statistical tests, we applied the q-value method (Storey and Tibshirani, 2003) with a threshold q<0.1 within each analysis. The carbonylation blots had fewer protein spots than the abundance gels, and no protein carbonylation spots met the q-value threshold for any of the three species. Lastly, to visualize proteome-wide treatment differences in abundance and carbonylation, we performed species-specific, unsupervised principal components analyses (PCA) and plotted individual mussels' PCA scores in 3D PC space.

We excised protein spots of interest using 1 mm biopsy punches, focusing on clearly identifiable spots that were most differentially expressed or contributed most to treatment differences in carbonylation. One set of protein spots were analyzed on an Applied Biosystems 4700 Proteomics Analyzer MALDI-TOF-TOF mass spectrometer at the University of California, Davis, USA, using previously described methods (Dowd et al., 2010a,b). Bioinformatics searches used both MS and MS/MS spectra to search against the NCBI and SwissProt Mollusca entries in Mascot (Perkins et al., 1999), with mass error tolerances of 50 ppm for MS fragments and 0.5 Da for MS/MS ions, and a maximum of 1 missed cleavage.

There were some examples of constitutive or inducible differences among the three mussel species in their antioxidant capacities and/or levels of oxidative damage, but these were not systematically different between M. trossulus and its more heat-tolerant congeners. Furthermore, repeated episodes of heat stress in seawater or in air had only muted effects on the antioxidant and oxidative damage profiles of the three congeners. However, there were several instances in which responses to heat stress in air differed dramatically from responses to heat stress in seawater.

We predicted a lower or less inducible activity of the antioxidant enzyme CAT in the heat-sensitive M. trossulus relative to either M. californianus or M. galloprovincialis. The activity of catalase was significantly impacted by species (P=0.002; Table S1), but not entirely in the predicted manner. As predicted, CAT activity of M. galloprovincialis exceeded that of M. trossulus in the control temperature seawater, SW1, Air1 and Air3 treatments. However, CAT activity of M. galloprovincialis also exceeded that of M. californianus in the control, SW1 and Air3 treatments (Fig. 1). CAT activity did not differ between heat-tolerant M. californianus and heat-sensitive M. trossulus. There was no overall treatment effect on CAT activity (P=0.933), but the interaction term was only marginally non-significant (P=0.064). For M. galloprovincialis, CAT activity after three episodes of heat stress in seawater was significantly reduced relative to that in all but the Air1 treatment (Fig. 1B; Table S1).

We also predicted a lower and/or less-inducible activity of the antioxidant SOD in M. trossulus. The data only partially supported this prediction. The linear model indicated significant effects of species (P<0.001), treatment (P=0.033) and a species-by-treatment interaction (P=0.017) (Table S2). The three species had statistically indistinguishable SOD activity in control conditions, but they responded differently to heat stress episodes (Fig. 2). SOD activity in M. galloprovincialis exceeded that of M. trossulus only in SW1. Similar to CAT activity, SOD activity in M. galloprovincialis exceeded that of M. californianus in SW1, Air1 and Air3. Contrary to the prediction, SOD activity in heat-tolerant M. californianus (Fig. 2A) and heat-sensitive M. trossulus (Fig. 2C) remained similar across the five treatments.

If accumulation of oxidative damage to lipids contributes to reduced heat tolerance, we would predict greater levels of this damage marker in heat-sensitive M. trossulus. There was a significant effect of species on levels of TBARS (P=0.002; Table S3), but not in the expected direction. On average, levels of lipid oxidation in heat-tolerant M. galloprovincialis were roughly 1.5–2 times higher than those in the other two congeners (Fig. 3). We confirmed the constitutive excess of TBARS in M. galloprovincialis relative to M. trossulus in an independent experiment (W.W.D., unpublished data). Furthermore, there was no accumulation of TBARS lipid peroxidation products in any of the species after 3 days of elevated body temperature in either air or seawater (P_treatment=0.442; P_{treatment×species}=0.411).

Limited nucleic acid material precluded replication of this assay, but using pooled samples, we found that gill 8-OHdG levels in heat-tolerant M. californianus were more than double those of both heat-sensitive M. trossulus and heat-tolerant M. galloprovincialis in control conditions. Levels of 8-OHdG in M. californianus remained higher in the seawater and air heat stress treatments (Fig. 4). There was little indication of accumulation of oxidative DNA damage following multiple episodes of heat stress in any of the three congeners. These results contradict the prediction that heat-sensitive M. trossulus should suffer more DNA oxidative damage.

We quantified protein abundance and signals of protein oxidative damage (amino acid side-chain carbonylation) in control conditions and after three episodes of heat stress in air and in seawater. We predicted that a greater number of proteins, and/or specific proteins important in thermal defenses, would suffer increased levels of carbonylation following repeated episodes of heat stress in the heat-sensitive congener M. trossulus. Both protein abundance gels and carbonylation blots were generally highly reproducible across biological replicates (Fig. S1). One M. trossulus abundance gel (in the seawater heat treatment) and one M. trossulus carbonylation blot (in the control treatment) were excluded from analyses because of poor protein separation and image quality.

The air and seawater heat treatments induced largely distinct changes in protein abundance in both M. galloprovincialis and M. trossulus (abundance data were not collected for M. californianus). A total of 51 of the 1340 protein spots on M. galloprovincialis abundance gels were significantly differentially expressed among at least two of the three treatments at P<0.01 (20 of these with a q-value<0.10) (Table 1; Table S4). Of the 51, only 17 proteins increased in abundance relative to controls in both heat treatments, and 6 proteins decreased in abundance in both heat treatments (including spot 31357 identified as 60 s ribosomal protein l13a; Table 2). A larger proportion of the differentially expressed proteins (33 of 51) varied between the two heat treatments. Notably, four protein spots for M. galloprovincialis were strongly upregulated in the seawater heat treatment and were identified with mass spectrometry as variants of Hsp70 or glucose-regulated protein 78 (Table 2); each of these chaperone proteins was approximately half as abundant after heat stress in air. Spot 124369, identified as M. edulis heavy metal binding protein (c1q domain containing), was similarly much more abundant after heat stress in seawater than in air (Table 2). There was clear treatment separation of protein abundance profiles within principal component (PC) space, particularly along the first dimension for which the repeated seawater heat stress treatment had elevated scores (Fig. 5B).

Of the 1374 spots on M. trossulus protein abundance gels, 57 were significantly differentially expressed in any comparison at P<0.01 (27 of these with a q-value<0.10) (Table 1; Table S4). More proteins increased in abundance in heat stress treatments relative to controls than decreased; 18 proteins were elevated in both seawater and air heat stress treatments (including 47924 Hps70; 47944 Hsp70-1; 47906 heat shock protein 70 B2-like; Table 2). Unlike for M. galloprovincialis, the M. trossulus proteins that changed significantly in both heat treatments tended to be more highly upregulated in air than in seawater (6 of 18 instances) or upregulated to equal degrees in the two heat treatments (10 of 18 proteins) (Table S4). Another 6 proteins decreased in abundance in both heat treatments relative to controls. As for M. galloprovincialis, a larger proportion of differentially expressed proteins (41 of 57) varied between the two heat treatments than differed between either heat treatment and the control. Accordingly, there was clear separation of proteome-wide abundance patterns among the three treatments in PC space, indicating distinct expression patterns in each treatment (Fig. 5D).

Overall, there was little evidence for induction of protein carbonylation by either repeated heat treatment. There was also little variation in overall sensitivity to this form of damage among the three congeners (Table 1). Instead, the data indicate a broadly similar pattern of protein carbonylation in control conditions and after repeated heat exposures for each Mytilus congener.

Only 2 of the 112 consistently detected carbonylated protein spots on M. californianus blots were significantly different in any comparison (P<0.01). One of these increased in signal relative to the control following seawater heat stress, and the other decreased in the same treatment (Table 1; Table S4). Because we did not have independent protein abundance gels from which to pick spots, the identity of these two proteins is unknown. The first three PCs explained 84.35% of the variance but revealed no separation of the treatments in PC space (Fig. 5A).

Similar to M. californianus, protein carbonyl signals for M. galloprovincialis were largely unchanged following air or seawater heat treatment. Only 10 of 507 carbonylated protein spots were significantly different in any comparison (P<0.01) (Table 1; Table S4). In 12 of the 15 total pairwise instances of significant changes, carbonylation signals of individual proteins decreased in the repeated heat stress treatments relative to the controls (Table S4). No protein carbonyl signals increased in both seawater and air heat treatments; 5 significantly decreased by similar amounts in both heat treatments (including spot IDs 67231, 70506 and 67341, the last of which was identified as actin A3; Table 2). Thus, there is no evidence of a significant rise in carbonylation. These carbonylation signal changes were independent of changes in protein abundance. In 9 of 10 cases, we were able to map the differentially carbonylated spots to a corresponding spot on protein abundance gels, and none of those 9 proteins changed significantly in abundance between any two treatments (Table S4). As for M. californianus, there was little evidence of treatment separation of proteome-wide carbonylation patterns in PC space for M. galloprovincialis (Fig. 5C).

The most heat-sensitive congener, M. trossulus, also exhibited very little change in protein carbonylation signals following either heat treatment. Only 2 of 422 carbonylated protein spots were significantly different in any comparison for M. trossulus (Table 1; Table S4). One spot (ID 46774) had 4.7- and 7.3-fold increases in carbonylation signal after seawater and air heat treatment, respectively. However, the matching spot on protein abundance gels (ID 47920) was 8.72 and 12.01 times more abundant in the heat treatments than in controls (Table S4). As a result, the ratio of carbonyl signal to total protein decreased for this spot in both heat treatments. Based on the location of this spot adjacent to M. trossulus ID 47924, which was identified as a Hsp70, we believe but have not confirmed that this protein is also a Hsp70. Mytilus galloprovincialis protein spots 30950 and 30954 were also found in the same gel positions and identified as Hsp70 variants. The other significant protein (ID 46705) had decreased carbonylation signal in the seawater heat treatment. In agreement with the few statistically significant changes and in line with the other congeners, there was no separation of the three treatments in PC space for M. trossulus (Fig. 5E).

Differential resistance to oxidative stress has received attention in the literature as a possible mediator of variation in thermal tolerance (Kassahn et al., 2009; Pörtner, 2002), including in the Mytilus mussel congeners studied here (Popovic and Riginos, 2020; Tomanek and Zuzow, 2010). Behavioral observations had also led us to speculate that M. trossulus' relative avoidance of periods of valve closure during recovery from episodes of heat stress was a means to avoid additional hypoxia–reoxygenation cycles and concomitant oxidative stress (Dowd and Somero, 2013). However, using repeated heat treatments that are known to induce differential mortality among the three species (Dowd and Somero, 2013), we found little evidence to support the hypothesis that oxidative damage contributes to differences in acute heat tolerance among three Mytilus congeners.

Species differences in these data are generally inconsistent with the hypothesis that oxidative stress limits thermal tolerance among Mytilus congeners, at least under the realistic, episodic exposure regimes we employed. The primary pieces of evidence in support of the hypothesis are higher (CAT) or more inducible (SOD) antioxidant capacities in heat-tolerant M. galloprovincialis than in heat-sensitive M. trossulus. Other work has revealed similar heat inducibility of antioxidant defenses in M. galloprovincialis (Georgoulis et al., 2021; Wang et al., 2018). However, these elevated antioxidant capacities, which are combined with the genomic features and expression differences described above, do not protect M. galloprovincialis from having equal or sometimes higher (in the case of TBARS) levels of oxidative macromolecular damage. This pattern is somewhat surprising, given the fact that M. trossulus has a faster aerobic ‘pace of life’ than M. galloprovincialis based on citrate synthase enzyme capacity (Lockwood and Somero, 2011; Fig. S2), which might be expected to lead to greater cumulative levels of ROS formation in M. trossulus. However, ROS formation and aerobic metabolism are often only weakly coupled (Koch et al., 2021; Monaghan et al., 2009). Our conclusion that oxidative stress is not a major contributor to differences in heat tolerance among Mytilus congeners is further supported by the results from the second relatively heat-tolerant species M. californianus, which exhibits antioxidant capacities on a par with those of heat-sensitive M. trossulus yet suffers similar (TBARS) or greater (8-OHdG) quantities of oxidative damage.

Furthermore, there was little evidence for a pronounced increase in any of the measures of oxidative damage to macromolecules, or for induction of antioxidant defenses, in the more thermally sensitive M. trossulus after exposure to repeated (three) episodes of heat stress in air or seawater. Again, our temperature treatments are known to induce substantial, but delayed, mortality in M. trossulus (Dowd and Somero, 2013). Thus, our data strongly suggest that accumulation of oxidative damage is not a key contributor to this mortality. Metrics of antioxidant capacity and oxidative damage also were little changed in the other two Mytilus congeners. Similar results indicating no effect of different thermal treatments on antioxidant parameters or total protein carbonyls were observed within a species of intertidal limpet (Drake et al., 2017). The 2D protein carbonyl blots similarly provide no quantitative evidence for a broad increase in oxidative protein damage in the heat-sensitive M. trossulus relative to its more heat-tolerant congeners (Table 1).

In lieu of a widespread increase in protein oxidation, we predicted that the individual proteins that are susceptible to heat-induced carbonylation in M. trossulus might be important in thermal defenses. Protein carbonylation is known to be limited to a subset of the proteome (Cabiscol et al., 2000), with the most susceptible proteins including those with catalytic motifs that contain transition metals as well as enzymes associated with energy production and conversion (Chaudhuri et al., 2006; Maisonneuve et al., 2009). For example, SOD and glyceraldehyde-3-phosphate dehydrogenase (GAPDH) were selectively carbonylated in yeast treated with hydrogen peroxide (Costa et al., 2002); GAPDH, aconitase and the alpha subunit of ATP synthase were carbonylated in yeast mutants lacking functional MnSOD (O'Brien et al., 2004). Other studies have similarly observed oxidative damage to protein components of the antioxidant system (Bruno et al., 2009), which could have the undesirable effect of dampening a cell's ability to defend itself against further oxidative damage (O'Brien et al., 2004).

The one protein in M. trossulus for which the carbonylation signal increased significantly after repeated episodes of heat stress in air and in seawater was putatively identified as the molecular chaperone hsp70. This raises the intriguing possibility that one of the most important proteins for maintenance of proteome integrity during heat stress is itself susceptible to heat-induced oxidative damage. However, the abundance of the corresponding protein spot increased more following heat stress than did the carbonylation signal, indicating that the relative amount of protein oxidation actually decreased following heat stress in this inducible chaperone. In vitro studies of purified M. trossulus hsp70 chaperone function before and after oxidation would be needed to evaluate the functional impacts of carbonylation. Other oxidative stress studies using similar methods have identified molecular chaperones as specific targets of carbonylation, including age-related damage to the protein-folding chaperone BiP in mammalian liver (Rabek et al., 2003), hsp70 in a copper toxicity study in yeast (Shanmuganathan et al., 2004) and hsp70 in a neurocognitive impairment model (Sultana et al., 2010).

Rather than implying a role for oxidative damage in differential heat tolerance, our protein abundance data confirm that air or seawater exposure to heat stress induces changes in protein expression patterns whose primary role may be to promote the conformational integrity of the proteome. Indeed, we observed strong upregulation of several proteins identified as heat shock proteins in both M. galloprovincialis and M. trossulus, as observed in previous Mytilus transcript and protein expression studies (Feidantsis et al., 2020; Fields et al., 2012; Lockwood et al., 2010; Tomanek and Zuzow, 2010). These results offer additional support for the notions that susceptibility to protein destabilization/denaturation is a primary consequence of heat stress and that variation in this susceptibility is a key driver of Mytilus species differences in heat tolerance (Gracey et al., 2008; Hofmann and Somero, 1995; Lockwood and Somero, 2012). Temperature effects on macromolecular stability loom large as a selective force in the evolution of thermal tolerance (Liao et al., 2021; Somero et al., 2017). While oxidative stress clearly plays a role in some aspects of life-history evolution and tradeoffs (Janssens and Stoks, 2018; Koch et al., 2021; Monaghan et al., 2009), its role in acute thermal tolerance remains less clear.

Despite considerable evidence that heat stress increases ROS production and/or oxidative damage in marine organisms (Abele et al., 1998, 2001; Heise et al., 2003; Jimenez et al., 2016; Wang et al., 2018), our measurements indicate a lack of pronounced changes in any of the oxidative damage metrics following repeated episodes of acute heat stress. We turn our attention to potential caveats to this observation.

First, it is possible that three ∼4 h episodes of acute heat stress constitute insufficient time for the manifestation of changes in the antioxidant system or in oxidative damage metrics. However, previous studies have observed abrupt changes (within 1 h) in oxidative damage metrics over single episodes (e.g. Bagnyukova et al., 2007; Wang et al., 2018), including acute rises in gill lipid hydroperoxides during heat exposures in M. californianus (Jimenez et al., 2016) and M. galloprovincialis (Wang et al., 2018). It may well be that chronic exposure to high temperature induces fundamentally different changes in these measures to those following the episodic regime we employed. For example, the activity of antioxidants including CAT and SOD increased and TBARS accumulated in response to chronic thermal stress (>5 days) in seawater in M. galloprovincialis (Feidantsis et al., 2020). However, in the region from which we sampled mussels, episodic exposure to heat stress during low-tide periods is the most likely scenario to challenge the thermal tolerance of these species.

Second, some oxidative damage products might be dynamically cleared during the ∼20 h period between the end of the final heat episode and our sampling time, precluding their detection in our design. For example, lipid oxidation products increased transiently and recovered within 24 h in a fish (Bagnyukova et al., 2007), and lipid peroxidation and DNA 8-OHdG damage accumulated in the mussel Perna perna when exposed to air but returned to baseline levels within 3 h of re-immersion (Almeida et al., 2005). In contrast, 8-OHdG peaked 24 h after reperfusion in a mammalian ischemia model (Liu et al., 2004). However, we observed no significant changes in TBARS over a shorter-term time series in heat-sensitive M. trossulus in seawater; we sampled at the peak of the heat ramp, immediately at the end of the ramp, and after 7 h of recovery (data not shown; n=5, P_{sampling-time}=0.541).

The situation for protein carbonylation is also complicated, in part because it is challenging to distinguish between carbonylation as a sign of damage (McDonagh and Sheehan, 2006) and carbonylation as the result of ROS-dependent signaling. Because they form via multiple mechanisms, protein carbonyls were long thought to be a comprehensive and stable index of oxidative protein damage (e.g. in disease states; Hecker and Wagner, 2018; Korolainen et al., 2007). However, reversal of protein carbonylation (decarbonylation) over short time scales of less than 30 min has been reported (Wong et al., 2008), putatively involving a thiol-dependent mechanism (Wong et al., 2013), although this process is not widely documented. Carbonylated proteins might also be degraded by the proteasome (Grune et al., 1997), preventing their accumulation. For example, we observed a decrease in carbonylation of an actin protein in M. galloprovincialis in both heat treatments, whereas in earlier work a mussel actin was deemed highly susceptible to carbonylation caused by pollution and hydrogen peroxide (McDonagh et al., 2005). Furthermore, carbonylation of specific amino acid side-chains is only one of many possible oxidative modifications of proteins, and some of these such as surface methionine oxidation are known to be reversible (Stadtman and Levine, 2003). There is some evidence that methionine oxidation and protein carbonylation may be linked (Moskovitz and Oien, 2010). Interestingly, in M. californianus, recent work suggested a peak in expression of a methionine-R-sulfoxide reductase protein during high tide (Elowe and Tomanek, 2021), perhaps indicative of a repair phase for methionine oxidation when oxygen and food are abundant.

With these caveats in mind, we cannot rule out transient rises and falls in the other measures of oxidative damage. Nonetheless, the near-complete absence of accumulation of any of the measured oxidative damage products after repeated episodes of thermal stress that are known to induce differential mortality strongly suggests that oxidative stress is not differentiating Mytilus congeners in their tolerance of acute heat stress. Instead, the intermittently warm and hypoxic lifestyle of all Mytilus mussels appears to have selected over evolutionary time either for resistance to ROS production and excess oxidative damage (Sokolov et al., 2021) or, as in the case of lipid oxidation, for favoring infrequent repair over the costly maintenance of a constitutive defense (Jimenez et al., 2016). To examine the generality of this conclusion that oxidative stress is not a major player in differential acute heat tolerance of Mytilus congeners, it will be necessary to gather similar comparative data from other genera that do not routinely experience episodic oxidative challenges in the same manner.

Our results support the assertion that air and seawater heat stress impose unique challenges for Mytilus mussels and other intertidal animals (Dowd and Somero, 2013). The distinction is particularly evident in the protein abundance results (Fig. 5B,D). Interestingly, these contextual differences are also most prominent in the more heat tolerant of the blue mussels, M. galloprovincialis, for which repeated exposure to heat stress in seawater is most distinct among the three treatments for global protein expression (Fig. 5B) and is associated with reduced antioxidant enzyme capacities (Figs 1 and 2). The same SW3 treatment induced the greatest increase in mortality for both M. galloprovincialis and M. trossulus (Dowd and Somero, 2013).

It is possible that these context-dependent differences in the response to heat stress have played a role in the invasive spread of M. galloprovincialis on the Pacific coast of North America. For example, our protein abundance patterns (including stronger induction of heat shock proteins after seawater heat stress than after air heat stress) and antioxidant data provide some functional evidence supporting the field observation that M. galloprovincialis fares worse in subtidal locations (Schneider and Helmuth, 2007). Interestingly, in other parts of the world, M. galloprovincialis has invaded intertidal habitats first (Robinson and Griffiths, 2002). Successful invasions arise from numerous contributions, including the genetic and physiological features of the populations' founding individuals and different suites of abiotic and ecological constraints in different regions. For example, prior work on the Pacific coast of North America has shown that the two blue mussels differ in their responses to the combined effects of temperature and salinity. The native M. trossulus is generally more successful in areas where salinity is lower on average, even if those habitats have warmer seawater temperatures (Braby and Somero, 2006a). It would be informative to repeat these experiments at a range of salinities to further examine the physiology underlying this interaction. Teasing apart the physiological consequences of such interactions remains a substantial challenge in the study of species invasions.

Lastly, the relationships between aerobic metabolic capacity, as indexed by citrate synthase activity, and either heat tolerance or susceptibility to acute oxidative damage are not straightforward among these Mytilus congeners. Mytilus californianus appears to have the highest aerobic pace of life of the three congeners studied here (Fig. S2), despite being the most heat tolerant of the three congeners we tested, with a median lethal temperature above 38°C in our source population (Denny et al., 2011). By some measures, M. californianus also suffers higher baseline levels of oxidative damage (8-OHdG) than the other congeners. However, lipid TBARS values are highest in M. galloprovincialis despite it having the lowest gill aerobic capacity. Interestingly, M. californianus evolved at a lower average temperature in the eastern Pacific (driven by the cool California current) than did M. galloprovincialis in the Mediterranean Sea. Thus, adaptation of aerobic and antioxidant enzymatic capacities may simply reflect long-term average oxidative rate, rather than predict the consequences of acute thermal stress.

In closing, we found little support for the hypothesis that accumulation of oxidative damage products contributes to higher mortality following acute episodes of heat stress in the mussel M. trossulus than in its more heat-tolerant congeners M. galloprovincialis and M. californianus. In several other respects, these mussels' antioxidant and oxidative damage profiles differ following exposure to heat stress in air versus in seawater, perhaps contributing to context-dependent patterns of invasiveness. Further comparative work in other genera is needed to examine the generality of these conclusions, given the expected increases in frequency and intensity of acute thermal stress events with climate change.

Laurie Kost collected citrate synthase data for M. californianus and coordinated the 2011 PISCO ‘Biomechanics and Ecological Physiology of Intertidal Communities' course for blue mussel genotyping. Jerod Sapp and Stephen Gosnell assisted with collection and shipment of bay mussels. Brent Lockwood provided unpublished data on M. galloprovincialis and M. trossulus gill citrate synthase activities. Mass spectrometry samples were processed in the laboratory of Dietmar Kültz at the University of California, Davis. Mark Denny provided constructive comments on the manuscript.