Deriving Individual Effect Sizes

To derive a quantitative understanding of ocean acidification effects on shellfish behavioral prey defenses, a meta-analysis was conducted. The effect size used was the natural logarithm-transformed response ratio (lnRR), which is defined as

where and are the average measured response in the experimental and control treatments, respectively. This effect size metric has a high capacity to detect true effects, is robust to low sample sizes, and quantifies the proportional change in a response variable due to experimental manipulations (Lajeunesse & Forbes 2003). It is important to note that, for some behaviors, an increase in the behavioral measurement corresponds to a negative functional response (but would result in a positive effect size estimate). In such instances, the directionality of the calculated effect size was inverted (i.e., positive values were changed to negative ones, and vice versa) to adequately reflect the functional consequence of the behavioral change.

Effect size variance was calculated as

where S and n are the SD and sample size, respectively, for a given experimental treatment (denoted by the subscripts C [control] and E [experimental, i.e., elevated pCO₂]); and are defined as aforementioned. Thus, studies with greater replication and less-variable responses were considered to be more precise estimates of a population effect and, as such, were weighted more heavily when calculating mean effect sizes (Hedges & Olkin 1985). This approach provides a conservative estimate of the mean effect size and increases the confidence that a statistically significant effect size is a true effect.

Determining Drivers of Individual Efect Sizes

Rather than assessing effect sizes between a priori defined groups, Akaikes information criterion (AIC) model selection was used to determine parameters that best accounted for the observed variability in effect sizes, and effect sizes were assessed across levels within each contributing parameter identified by AIC. Linear mixed models with additive fixed factors (i.e., interactions were not tested) were built for all possible combinations of predictor variables. Categorical predictor variables included in the models were taxonomy (phylum and class), climatic region (temperate, tropical, or polar; if animals were from the deep sea, they were considered in a separate group), and life stage (larvae, juvenile, or adult); study was included as a random variable to account for nonindependence due to the incorporation of multiple effect sizes from a single study. Before analysis, correlations between effect size and three continuous variables—exposure time, sample size, and pCO₂ offset (experimental–control pCO₂)—were also explored. There was a significant correlation between effect size and pCO₂ offset (r = –0.66, P < 0.001; Fig. 1), but not exposure time (r = –0.11, P = 0.377) or sample size (r = 0.16, P = 0.219); therefore, pCO₂ offset, but not exposure time or sample size, was included in the AIC models as a continuous predictor variable. Predictor variables were not significantly correlated (r < 0.36 for all pairwise correlations) with the exception of phylum and class (r = 0.93); therefore, models including both phylum and class as predictor variables were excluded from AIC selection. A “null” model (i.e., model with no fixed predictor variables and only a random factor of measurement nested within study) was also included in AIC selection. This resulted in a total of 24 additive models. Model selection was based on AIC and log likelihood values (Burnham & Anderson 2002; lowest AIC and highest log likelihood values indicate the “best” model), and all models with a delta AIC (Δ AIC) ≤2 were considered as “best” models (Burnham & Anderson, 2002). The AIC model selection was conducted on all studies, as well as only those with a pCO₂ offset consistent with near-future projections (<1,000 μatm), to determine whether predictors of effect size differed when pCO₂ projections beyond near-future were included versus excluded.

Effect Size Analysis of Predictor-Level Responses

Once the “best” model was defined, the mean lnRR and 95% confidence interval (CI) for each level of the predictor variable(s) that were present in the best model were calculated. To provide conservative estimates of effect size variance, bootstrapped (10,000 replicates), bias-corrected, and accelerated (BCa) 95% CIs were used (Adams et al. 1997). Mean effect sizes were estimated using weighted random effect models, which weight individual effect sizes by the inverse of the effect size variance (Hedges & Olkin 1985). Random effects models were used as opposed to fixed effects models because of inherent differences in conditions between studies (i.e., species and geographic location). To account for nonindependence associated with using multiple data points per study, mean effect sizes were calculated using three-level meta-analytical models (Nakagawa et al. 2015, Noble et al. 2017), including “measure nested within study” as a random variable. Mean lnRR and associated variance (bootstrapped BCa 95% CIs) were only derived for factor levels with n ≥ 3.

Figure 1.

Effect of pCO₂ on effect sizes. Absolute effect sizes as a function of pCO₂ offset (experimental–control pCO₂) with all pCO₂ anomalies (black plot) and only near-future pCO₂ anomalies (gray plots; pCO₂ offset <1,100 based on IPCC 2100 RCP8.5 projection) included.

Statistical significance was determined by assessing whether the BCa 95% CI surrounding a mean effect size overlapped with 0 (Lajeunesse & Forbes 2003). Heterogeneity in effect sizes for each level of the factors identified in the top candidate model(s) was tested using Q-tests of heterogeneity (Lajeunesse & Forbes 2003). Given the significant correlation between effect size and pCO₂ offset for all pCO₂ conditions (r = –0.66, P < 0.001), but the lack of correlation for near-future only (r = –0.28, P = 0.075), two separate analyses were conducted: one for all pCO₂ conditions and another for only near-future pCO₂ conditions (i.e., pCO₂ offset <1,000 μatm; upper estimate for average pCO₂ increase based on the IPCC RCP8.5 projection for 2100). A null effect was signified when lnRR = 0, whereas negative and positive effects were defined by negative and positive lnRR values, respectively (Hedges et al. 1999).

Sensitivity Analysis and Publication Bias

The influence of single studies that elicited unusually large effect sizes was investigated using sensitivity analysis. Herein, observations within each level of the factors present in the best AIC model were systematically ranked according to their observed mean effect size; the largest effect size was then removed from the dataset, and the analysis was rerun. The relative contribution of individual studies to the observed mean effect sizes was also considered; however, no single study contributed more than four experiments to the meta-analysis (three studies contained four experiments, whereas many other studies contained three experiments) and, as such, the data were not reanalyzed excluding studies that contributed multiple observations to the overall effect size. To test for publication bias, the number of studies with an lnRR = 0 that would be required to change a significant effect size response to a nonsignificant response was determined using Rosenthals fail-safe number (Rosenthal 1979). The analysis for a given category was considered robust if Rosenthals fail-safe number was ≥5 (Kroeker et al. 2010, 2013; Harvey et al. 2013).

Analytical Details

All analyses were conducted in R v. 3.4.1 (R Core Team 2017) with a significance threshold of α = 0.05. Effect size analyses were conducted using the metafor package (Viechtbauer 2010, 2017), with 95% BCa CIs generated using the “boot” package (Canty & Ripley 2017). Pearson correlations were tested using the rcorr function in the Hmisc package (Harrell Jr. 2017), AIC model selection analysis was conducted using the AICcmodavg package (Mazerolle 2017), and R² values for linear mixed models were computed using the MuMIn package (Barton′ 2018). Before analyses including individual effect sizes (i.e., Pearson correlations and AIC), effect sizes were transformed (ln+10) to achieve normality.

Meta-Analysis Results

Effects of ocean acidification on shellfish behavioral defenses were predominantly driven by pCO₂ offset, taxa, and life stage. When all pCO₂ conditions were considered, AIC model selection revealed the model including pCO₂ offset + class + life stage as the “best” model (Table 2). This model explained 59% (R² = 0.59) of the variance in individual effect sizes. A second model including pCO₂ offset + life stage could not be excluded (delta AIC = 1.75) and explained 58% (R² = 0.58). When only near-future pCO₂ anomalies where considered, a number of candidate models (i.e., those with a delta AIC <2) were evident; these included the model including pCO₂ offset, the model including class, the model including life stage, and the null model. Of these models, the model including class explained the most variation (R² = 0.30), followed by life stage (R² = 0.23), the null model (R² = 0.22), and pCO₂ offset (R² = 0.20). For both pCO₂ groupings, effect size was negatively related to pCO₂ offset, although the relationship for near-future pCO₂ values was weak and non-significant (Fig. 1). Given that the fixed factors of class and life stage appeared in the best models, the effect size analysis was restricted to testing for ocean acidification effects on prey defense behaviors across classes and life stages. It is important to note, however, that the R² values for all pCO₂ conditions were primarily explained by pCO₂ offset and that those for near-future pCO₂ only were relatively low. This suggests that the predictive power of class and life stage was relatively low; nonetheless, they appeared to be more important in explaining the effect size variation than other factors based on AIC model selection.

When all pCO₂ offsets were included, meta-analysis revealed that mean effect sizes of bivalve and malacostracan defenses were significantly lower than zero (Fig. 3A). By contrast, the mean effect size in cephalopods was significantly higher than zero, despite substantial variability (Fig. 3A). Effect sizes in both gastropods and echinoids were not significantly different from zero (Fig. 3A). Sensitivity analysis revealed that unusually large effect sizes did not influence the statistical outcome of class effect sizes (data not shown). The categorical analysis was robust (Rosenthals fail-safe >7) for all classes, except for echinoids [Table 3; Rosenthals fail-safe = 0, which was unsurprising given that all studies on echinoids reported exclusively null statistical effects (Fig. 2B)]. The significant negative effect size for bivalves was retained when only near-future pCO₂ conditions were considered (Fig. 3B). Likewise, the significant positive effect size for cephalopods remained when only near-future conditions were considered, along with the null effect size for gastropods and echinoids (Fig. 3B). In contrast to when all pCO₂ conditions were considered, the effect size for malacostracan crustaceans was not significantly different from zero (Fig. 3B) and the robustness of the malacostracan effect was reduced when only near-future conditions were considered (Table 3; Rosenthals fail-safe number = 2). Heterogeneity (Q-statistic) for all classes was significant for both pCO₂ groupings (Table 3).

TABLE 1.

Summary of experimental studies assessing the effects of elevated pCO₂ on the behavioral prey defenses.

continued

continued

Figure 2.

General overview of studies identified in this review. (A) The annual (gray plots) and cumulative (black plots) number of studies assessing the effects of elevated pCO₂ on behavioral defenses of shellfish prey from 2007 to present. Open plots represent the number of studies published during the first third of 2018 (i.e., January–April). (B) Taxonomic spread (pie chart) and statistical effects (bar plots) of the studies. Pie slices represent the percentage of studies that assessed particular taxa (classes: bivalves [black], cephalopods [dark gray], gastropods [gray], echinoids [light gray], and malacostracans [white]). Values within each pie slice denote the percentage of studies within a given taxa that tested one of three life stages: larvae (L), juvenile (J), or adult (A) (NR$ life stage not reported). Bar plots represent the number of studies within a given taxa that reported statistically negative (diagonally hatched), positive (vertically hatched), null (horizontally hatched), or mixed (dotted) effects. “Negative,” “positive,” and “null” were assigned when all reported effects of elevated pCO₂ within a given study were negative, positive, or null; “mixed” was assigned when responses to elevated pCO₂ varied within a given study (contingent on some variable factors such as pCO2 level, species, or behavioral metric).

TABLE 2.

Results of AIC model selection analysis for additive models explaining variance in individual effect sizes.

With respect to life stage, the meta-analysis revealed that only juvenile effect sizes were significantly lower than zero (Fig. 4). This was true for all pCO₂ conditions and near-future conditions only. The mean effect size for both larvae and adults was not significantly different from zero for both pCO₂ groupings (Fig. 4). Heterogeneity (Q-statistic) was significant for all life stages under both pCO₂ groupings (Table 3).

Figure 3.

Responses of shellfish behavioral prey defenses to ocean acidification across taxonomic class. Mean effect sizes (lnRR ± bootstrapped BCa 95% CIs) for five classes of shellfish: malacostracans (white), echinoids (light gray), gastropods (gray), cephalopods (dark gray), and bivalves (black), with all pCO₂ anomalies included (A) and only near-future pCO₂ anomalies included (B). Values in parentheses indicate the number of observations used to derive the mean effect size for each taxon. Asterisks denote significant responses (i.e., when 95% CI does not overlap with 0). Negative and positive values correspond to negative and positive functional effects, respectively.

Meta-Analysis Results in a Broader Context

Taxonomic class consistently appeared in the best AIC models as a predictor of effect sizes. Although previous meta-analyses on biological responses to ocean acidification include taxonomy as an explanatory variable, those studies typically select phylum as the taxonomic resolution at which to test differences (Kroeker et al. 2010, 2013, Harvey et al. 2013). The results of the meta-analysis suggest that, at least for calcifying organisms, this level of taxonomy may be too broad and that using finer scales of taxonomic resolution may provide better predications of ocean acidification sensitivity. It is important to note, however, that class alone was unable to explain much of the variation in observed effect sizes herein (R² = 0.27 for near-future pCO₂), and as such may not be a fine enough taxonomic scale to explain observed differences. Such results highlight the importance of species-specific biology. Furthermore, Q-tests of heterogeneity revealed significant variation across studies within each taxonomic class (Table 3), suggesting that other processes are certainly involved in explaining ocean acidification effects on behavioral defenses of shellfish reported in the experimental studies reviewed here. Such variation could be explained by some variable(s) not included in the analysis (such as the type of behavior, the biological state of the test animals, local adaptation, or mismatches between experimental exposures and natural conditions), and may largely reflect unaccounted experimental artifacts that differed across studies.

TABLE 3.

Results of effect size analysis, including mean effect size, bootstrapped 95% CIs, Q-test results, and Rosenberg fail-safe test results.

When all pCO₂ conditions were considered, a significant negative effect size was observed for the behavioral defenses of malacostracan and bivalve prey, whereas effect sizes for gastropods and echinoderms were not significantly different from zero; in addition, cephalopods exhibited a significant positive effect size. These results are in contrast with previous meta-analyses reporting mostly null effects of elevated pCO₂ on crustacean taxa and sensitivities in molluscs and echinoderms (Kroeker et al. 2010, 2013, Harvey et al. 2013, Whittman & Pörtner 2013). Previous meta-analyses have excluded metrics of behavior for shellfish taxa likely because too few studies were available (see Fig. 2A). The results of this study suggest that impacts of ocean acidification on other biological processes of shellfish such as physiology and survival may not be adequate in defining effects on shellfish behaviors. It is important to note, however, that the negative effect sizes for malacostracans observed in this study were driven by higher-than-projected pCO₂ conditions, as significant negative effects on malacostracans became nonsignificant when only near-future pCO₂ changes were considered. More confident predictions of shellfish behavioral defenses under near-future pCO₂ changes across classes, however, await further investigation, as sample sizes herein were relatively low (i.e., meta-analysis with only near-future pCO₂ conditions were not robust; see Rosenthal fail-safe numbers in Table 3).

Figure 4.

Responses of shellfish behavioral prey defenses to ocean acidification across life stage. Mean effect sizes (lnRR ± bootstrapped BCa 95% CIs) for three life stages of shellfish: adults (white), juveniles (gray), and larvae (black), with all pCO₂ anomalies included (A) and only near-future pCO₂ anomalies included (B). Values in parentheses indicate the number of observations used to derive the mean effect size for each life stage. Asterisks denote significant responses (i.e., when 95% CI does not overlap with 0). Negative and positive values correspond to negative and positive functional effects, respectively.

Life stage was also a fixed factor that consistently appeared in the best models based on AIC model selection. Effect size analysis revealed that juvenile mean effect sizes were significantly lower than zero when all and only near-future pCO₂ conditions were considered; larval and adult effect sizes were not significantly different from zero. Previous meta-analyses also highlight the importance of life stage (e.g., Kroeker et al. 2010, 2013, Harvey et al. 2013); however, those previous studies typically report that earlier life stages (i.e., larvae and juveniles) are more susceptible to ocean acidification than later life stages. By contrast, the meta-analysis detected no significant effect on larvae. Although this may suggest that larvae are robust to ocean acidification behaviorally, it is important to note that the number of studies on larvae was low (n = 6) and the nonsignificant effect size may simply reflect a lack of ability to detect an effect.

Although negative effect sizes under elevated pCO₂ in this study were largely driven by pCO₂ values exceeding near-future (i.e., year 2100) projections, such drastic changes in pCO₂ are not uncommon in nearshore coastal and estuarine waters where many of the species assessed herein reside. Diel and seasonal variation in coastal pCO₂ conditions can often far exceed near-future projections (Provoost et al. 2010, Duarte et al. 2013, Waldbusser & Salisbury 2014, Wallace et al. 2014). Furthermore, given that ocean acidification studies typically acclimate animals under control and elevated pCO₂ conditions for relatively short periods of time (compared with the large timescales across multiple organismal generations in which true ocean acidification will occur; McElhaney 2017), the effect sizes herein are mostly applicable to acute exposures to elevated pCO₂ characteristic of nearshore systems. The analysis thus suggests that shellfish behavioral defenses in nearshore coastal systems may already be affected by acute increases in seawater pCO₂. Given that the magnitude and temporal duration of pCO₂ extremes are expected to increase under ocean acidification (Kwiatkowski & Orr 2018, Pacella et al. 2018), the results herein suggest that shellfish behavioral defenses for some taxa are likely vulnerable to pCO₂ variability under current conditions and near-future ocean acidification, particularly when pCO₂ conditions are elevated for periods of weeks to months.

Interpretive Caution

It is important to note for this section that this meta-analysis is based on a relatively small number of studies. Given the high degree of heterogeneity observed (i.e., Q-statistics; Table 3), the small number of studies makes definitive conclusions regarding effects difficult. Consequently, the null effects reported here may represent “true” null effects or may be a symptom of low sample sizes (i.e., an inability to detect an effect). Thus, the meta-analysis herein provides a quantitative synthesis of studies to date, and definitive conclusions regarding effects await further research.

Multiple Stressors

Although some information pertaining to the interactive effects of ocean acidification and warming on shellfish behavioral defenses was gleaned, the combined effects of ocean acidification and other global change stressors are understudied. Of the 34 studies identified here, only one assessed the effects of ocean acidification in the context of a stressor other than temperature: Zhang et al. (2014) assessed the combined effects of pCO₂, temperature, and salinity on the swimming performance of two gastropods, Nassarius festivus and Nassarius conoidalis. Therein, it was reported that elevated pCO₂ reduced swimming speed in N. conoidalis (but not in N. festivus) and that higher temperature and higher salinity increased the swimming speed for both species.

Despite the lack of multistressor studies on shellfish behavioral defenses, it is widely known that responses to ocean acidification can depend on other stressors such as temperature, hypoxia and deoxygenation, salinity, and eutrophication, which can act synergistically or antagonistically with ocean acidification to drive biological responses (Pörtner 2008, Wallace et al. 2014, Breitburg et al. 2015). Food web structure and food supply are also considered to have impacts on biological responses to ocean acidification (Alsterberg et al. 2013, Ramajo et al. 2015, Goldenberg et al. 2018, Sswat et al. 2018). Adding to such complexities, ocean acidification can affect organismal affinity and selection for particular environmental conditions (e.g., temperature and salinity preferences as per Pistevos et al. 2017). Although not assessed in shellfish, studies have assessed ocean acidification effects in the context of some additional stressors on antipredator behaviors in fishes. For example, McMahon et al. (2018) reported that food availability had no influence on ocean acidification–induced effects on antipredator behavior in the juvenile clownfish Amphprion percula. By contrast, Miller et al. (2016) found that ocean acidification exposure led to increased surface ventilating behavior under low oxygen in congeneric fishes Menidia menidia and Menidia beryllina, which could potentially lead to increased vulnerability to aerial predators. Such complex effects highlight the importance of considering multiple global change stressors in predicting how the behavioral prey defenses of marine fauna will be impacted in a future ocean. When it comes to shellfish, more research is necessary, as research has yet to even scrape that surface.

Environmental Variability

As mentioned previously, coastal systems are often characterized by a high degree of spatial and temporal carbonate system variability (Provoost et al. 2010, Duarte et al. 2013, Waldbusser & Salisbury 2014, Wallace et al. 2014), which is likely to increase in the future (Kwiatkowski & Orr 2018, Pacella et al. 2018). From an organismal perspective, such variability is far from benign and can influence biological responses to ocean acidification in a number of potential ways (Helmuth et al. 2014, Waldbusser & Salisbury 2014, Bates et al. 2018). For example, a high degree of abiotic variability may confer tolerance to a wide range of environmental conditions in some coastal organisms, rendering them robust to relatively small baseline changes over long periods of time (i.e., many generations). Consequently, in coastal marine systems, pCO₂ variability is an important consideration in understanding biological responses to ocean acidification (Waldbusser & Salisbury 2014).

In the context of shellfish behavioral defenses, two of the 34 studies identified in this review assessed ocean acidification effects in the context of pCO₂ variability, reporting conflicting results. In a pioneering study, Alenius and Munguia (2012) showed that variable high pCO₂ conditions resulted in negative effects to locomotory behavior in the adult isopod Paradella dianae, whereas stable high pCO₂ conditions had no effect. By contrast, Jellison et al. (2016) reported no difference between stable and fluctuating CO₂ conditions on predator avoidance behaviors in the adult snail Tegula funebralis. Although the paucity of information regarding the role of pCO₂ variability (along with variability in other environmental parameters) in shaping the responses of shellfish behavioral defenses precludes firm conclusions, the contrasting results observed here underscore the importance and complexity of such variability.

Evolutionary Effects

An increasingly popular topic in contemporary ocean acidification research concerns the role of evolution and plasticity in biological responses to ocean acidification (Sunday et al. 2014, Calosi et al. 2016). Evolutionary forces such as local adaptation (Vargas et al. 2017) and natural selection (Thomsen et al. 2017), along with acclimatory processes such as transgenerational plasticity (Parker et al. 2012, 2015, Borges et al. 2018), can play important roles in shellfish responses to ocean acidification, both in isolation and in the presence of other co-occurring stressors (Gibben et al. 2017, Griffith & Gobler 2017). The literature search revealed no articles testing for evolutionary or acclimatory effects on shellfish behavioral defense responses to ocean acidification. Furthermore, a broader search of the literature suggested that evolutionary studies for invertebrate behaviors (in a broad sense) under ocean acidification remain absent (Ross et al. 2016), with the exception of a recent study reporting that transgenerational acclimation did not alleviate ocean acidification effects on mate guarding in the amphipod Gammarus locusta (males guard females from competing mates) (Borges et al. 2018).

Although evolutionary studies of shellfish behavioral defenses are not yet available, studies to this regard in fish may shed some light on potential effects in shellfish. For example, Welch et al. (2014) observed no evidence that transgenerational acclimation affected offspring behavioral responses (behavioral lateralization) to ocean acidification conditions in Acanthochromis polyacanthus. In a follow-up study on the same species, Welch and Munday (2017) demonstrated that although acute exposure to elevated pCO₂ elicited heritable behavioral phenotypes in response to predators, prolonged (four week) exposure to elevated pCO₂ negated the observed heritability under acute exposure. Ultimately, the influences of evolution and transgenerational plasticity on behavioral responses of shellfish to ocean acidification deserve attention, particularly given the variable impacts that these processes can have on shellfish responses to ocean acidification (Ross et al. 2016).

Ecological Complexity

The experiments reviewed herein suggest that ocean acidification can potentially influence behavioral defenses in shellfish prey, based primarily on laboratory observations of individual behaviors under control and elevated pCO₂ conditions. Predicting how and whether such changes might manifest themselves under natural conditions is difficult, however, because of the inherent situational and ecological complexity of nature (i.e., the breadth of other things that are occurring in natural systems). For example, individual personalities within a group of organisms can affect group-level responses to predation (Briffa et al. 2008). Similarly, immediate and prior exposures to the risk of predation (e.g., through predator-produced kairomones or through conspecific alarm cues) can influence behavioral prey defenses in shellfish (Hagen et al. 2002, Briffa 2013). Strikingly, of the studies reviewed herein, only 15% (n = 5 of 34) included predator cues in their experiments testing for effects of ocean acidification on prey defense behaviors (Bibby et al. 2007, Hatfield-Vaughan 2011, Manríquez et al. 2014, Watson et al. 2014, Jellison et al. 2016). This important situational caveat calls into question the transferability of many of these experiments to natural conditions. Studies including predator cues, however, have still reported significant ocean acidification effects on shellfish defense behaviors, suggesting that behavioral alterations can still affect predator–prey ecology in systems involving shellfish prey (Manríquez et al. 2014, Watson et al. 2014, Jellison et al. 2016).

Other nuances of natural biological communities can likely influence shellfish behavioral defense responses to ocean acidification as well. Recent studies have suggested that although ocean acidification may directly affect individual behaviors of organisms, increased diversity within biological systems may allow for compensatory processes to offset negative direct behavioral effects related to consumer foraging (Goldenberg et al. 2018; although they did note that increases in risky behavior under ocean acidification may make prey more vulnerable to predation). Ultimately, the situational conditions within complex ecological systems may impact shellfish behavioral defenses against predation in a high CO₂ ocean and thus warrant exploration. Furthermore, an understanding of how such behavioral changes influence ecological structure and function is required to understand and predict the ecological consequences of ocean acidification.

Functional Trade-offs

Engaging (or not engaging) in a particular defense behavior may not only affect the vulnerability of an organism to predation but can also result in functional trade-offs with other biological processes. For example, Bridger et al. (2015) reported that bolder male hermit crabs Pagurus bernhardus experience lower fecundity than less bold males. Consequently, if ocean acidification results in increased boldness and risk-taking behavior in a given species, not only would vulnerability to predation increase but fecundity may also decrease. Such additive negative effects could have major implications for the ecological functioning of biological communities. These interactions between biological processes may not always be additively negative, however. For example, although reduced burrowing capacity under sediment acidification (Clements & Hunt 2017) conditions can make bivalves more vulnerable to predation (Flynn & Smee 2010), this behavioral response also allows bivalves to avoid other mortality sources such as shell dissolution (Green et al. 2004, 2009). Ultimately, the complex trade-offs between defense behaviors and other biological processes that drive net fitness outcomes and the capacity of organisms to survive and persist under ocean acidification deserve attention.