Megalopae were collected during the NOAA WCOA cruise in May and June of 2016 ( Figure 1 ; Table 1 ). Neuston and Bongo nets with a mesh size of 333 mm were deployed obliquely across a depth of ~60m, which spans the vertical habitat of larval Dungeness crabs (Hobbs and Botsford, 1992; Rasmuson, 2013) for a duration of approximately 20 minutes. Megalopae were sampled from tow samples and stored in 100% non-denatured ethanol for later analysis. The distance to shore for each station was estimated using the closest 0.04° gridbox from the NASA Ocean Biology Processing Group’s product (https://oceancolor.gsfc.nasa.gov/docs/distfromcoast/).

Complementary water column variables were collected across vertical profiles at each site including: temperature (T), salinity (S), DO, nitrate ( NO 3 − ) and phosphate ( PO 4 2 − ) concentrations, total dissolved inorganic carbon (TC), alkalinity and pH (measured at 25°C - corrected to in situ temperature, and expressed on the total pH scale). Subsequently, seawater calcite saturation state (Ω_c) was calculated using CO2SYS as described by Feely et al. (2016). At most stations, water data exist at approximately 10-m intervals over at least the upper 60 m approximating megalope habitat. However, the maximum water sampling depth at nearshore stations 114 and 130 was 35 and 41 m, respectively.

Megalopae were scanned using a Bruker Skyscan 1173 within the Karel F. Liem Bioimaging Facility at the University of Washington’s Friday Harbor Laboratories. While μCT could be used to measure a variety of exoskeleton metrics (e.g. volume, shell thickness, etc.), our method was optimized for the high throughput analysis of exoskeleton density with the hope that such measures become a routine component of marine monitoring. Between 2 and 4 megalopae were loaded dry into a plastic 2 ml vial and scanned simultaneously in air without a filter at a voltage of 49 kV, current of 160 mA and 0.3° rotation steps. Total time for each scan was less than 30 minutes, or approximately 10 minutes per megalope. The raw output of each scan was a histogram of greyscale values from 0 (black) to 255 (white), for which higher values indicate greater density. Scans from the same station were aggregated into an average histogram for that site representing 7 to 13 specimens ( Table 1 ).

The megalope greyscale signal in each histogram is convolved with the background signal of the plastic vial, which must be removed ( Figure 2 ). To achieve this, we identified the first greyscale value whose proportion increases from the previous value as the lower limit of the megalope signal and the first subsequent grayscale value whose proportion is<0.1% of the total as the upper limit ( Figure 2 ). We then integrated between these values to determine a final greyscale value for each site. Given that the exoskeleton of megalopae are likely a complex mixture of calcite, amorphous calcium carbonate, and chitin (Luquet and Marin, 2004; Boßelmann et al., 2007), with densities of 2.71 g/cm³, 1.62 g/cm³ (Bolze et al., 2002), and 1.4 g/cm³(Carlstrom, 1957), respectively, it is difficult to produce appropriate standards with which to convert greyscale values to absolute density. Furthermore, while higher greyscale values generally reflect denser material, the abundance of exoskeleton pores smaller than the ~20 mm resolution of our method (i.e., micropores) can complicate this relationship. To compute density from a μCT scan, it is necessary to identify a threshold value that separates mineral or exoskeleton from air (Meng et al., 2018; Fordyce et al., 2020). A high abundance of micropores increases the number of voxels that are not entirely exoskeleton or air; a phenomenon known as ‘partial voxel filling’ (Fordyce et al., 2020). The higher the frequency of partial voxel filling, the greater the likelihood that voxels of mostly air with a small proportion of high-density exoskeleton are identified as entirely low-density exoskeleton. Given these complications, we do not attempt to calibrate μCT data to absolute density, and instead present data in native greyscale units.

The environmental variables that may control differences in megalope exoskeleton density were evaluated by comparing density to a suite of plausible predictors that represent realistic megalope histories prior to collection. These scenarios are summarized in Table 2 . In the simplest scenarios (int30 and int60), it was assumed that the recovered specimens molted from z5 zoeae near the point of collection such that the environmental conditions measured at the time of sampling were representative of their entire history as megalopae. In these scenarios we considered vertical migrations over either the upper 30 or 60 m, which is broadly consistent with field observations (Hobbs and Botsford, 1992; Rasmuson, 2013). The predictor pool for these scenarios included T, S, DO, NO 3 − , PO 4 2 − , TC, alkalinity, pH and Ω_c averaged over depths of either 30 m or 60 m. For select instances where a station’s maximum depth was less than 60 m, data at the maximum available depth was used. To consider particular sensitivity to extreme values, we also considered the maximum and minimum of the variables over either the upper 30 or 60 m. Given recent work suggesting that megalope carapace dissolution is sensitive to vertical gradients in Ω_c (Bednaršek et al., 2020), we also added such gradients for T, S and Ω_c (ΔT, ΔS, ΔΩ_c), where the Δ-value is the difference between the parameter at the surface and 30 m or 60 m.

In a second set of scenarios (DVM30_xx and DVM60_xx), we considered the possibility that megalopae molted from z5 zoeae up to 30 days prior to collection, and experienced environmental conditions different from in situ measurements as they migrated to the sampling site. Because megalopae are active swimmers, we considered diel vertical migrations over the upper 30 m in DVM30_xx scenarios and over the upper 60 m in DVM60_xx scenarios. The total time between z5-megalope molt and subsequent megalope-juvenile molt has been previously estimated at 30 days (Norton et al., 2020), but the exact time between z5-megalope molt and collection for the specimens we analyzed is unknown. Because of this uncertainty we performed sensitivity tests that assumed z5-megalope molt occurred 10, 20 or 30 days prior to collection. These scenarios are distinguished with a subscript, such that DVM30₁₀ identifies the scenario where the z5-megalope molt occurred 10 days prior to collection, after which megalope migrate over the upper 30 m of the water column to the site of collection.

To evaluate these scenarios we leveraged a series of existing experiments that used JISAO’s Seasonal Coastal Ocean Prediction of the Ecosystem (J-SCOPE) model to explore how exposure histories influence the presence or absence of M. magister (Norton et al., 2020). J-SCOPE is a high-resolution biogeochemical model of Washington-Oregon shelf waters with demonstrable predictive skill on seasonal timescales (Siedlecki et al., 2016), which has been extensively evaluated against historical conditions (Malick et al., 2020; Norton et al., 2020). Running the model “in reverse” using hourly historical simulations allows a megalope’s trajectory to the site of collection to be estimated, and, in turn, record the ocean conditions that coincide with that trajectory (Norton et al., 2020). One hundred particles for each sampling station were considered and the mean of the environmental conditions experienced by the particles was taken as the best estimate for a site. Diel vertical migrations were incorporated into back-tracking experiments, assuming a maximum depth of 30 m (DVM30_xx; Figure 3 ) or 60 m (DVM60_xx; Figure 4 ). The back-tracked trajectories include both vertical and horizontal random walk terms based on diffusion. The vertical component is calculated using the time-variable, hourly stored 3D values of vertical diffusivity, while the horizontal component assumes a constant background diffusivity value of 1.0 m²/s. These terms ensure dispersion of particles released from a shared location. The following variables estimated by J-SCOPE were considered as predictors of greyscale density: T, S, DO, NO 3 − , pH, Ω_c, phytoplankton concentration ([phyt]), and their maximum and minimum values, plus ΔT, ΔS and ΔΩ_c. Maximum and minimum values were defined as the average of the highest and lowest 5% of values for each variable. Following Norton et al. (2020), we also calculated severity indices (SI: Hauri et al., 2013; Bednaršek et al., 2017) for T, S, DO, pH and Ω_c. Stations 127 and 130 fall outside of the J-SCOPE domain, so comparisons to DVM behaviors were limited to the remaining seven stations.

In a third set of scenarios (molt30_xx and molt60_xx), we modeled the possibility that megalope exoskeleton density was determined only by environmental conditions during the short period surrounding the z5-megalope molt, 10, 20 or 30 days prior to collection. To do so, we considered only the final 24 hours of the DVM30_xx and DVM60_xx exposure history simulations described above. We considered all depths to be equally likely for the z5-megalope molt and calculated average values of T, S, DO, NO 3 − , pH, Ω_c, [phyt], and their maximum and minimum values. We did not consider vertical gradients or severity indices in this set of scenarios. As above, these scenarios were limited to the seven stations within the J-SCOPE domain.

Our final scenarios (all30_xx and all60_xx) considered the possibility that the scenarios above are not mutually exclusive. For example, conditions at the time of z5-megalope molt and the exposure history between molt and collection could both influence density. Thus, we combined the predictors of the three other scenarios at each depth (e.g. int30, DVM30_xx and molt30_xx) for the seven stations within the J-SCOPE domain into a single pool.

For each scenario, we used a forward stepwise regression approach (Wilks, 2006; Saenger and Evans, 2019), in which the Akaike Information Criterion (Akaike, 1974) modified for small samples (AICc; Burnham and Anderson, 2010; Burnham et al., 2011), served as a stop criterion. Specifically, univariate regressions between μCT greyscale and all potential predictors were calculated, and the regression with the lowest AICc was retained. Subsequently, all bivariate regressions were calculated. If a bivariate model had a still lower AICc, the additional variable was retained, and the process continued. If no bivariate model had a lower AICc, the univariate model was considered to be optimal. Given the absence of data to the contrary, we assume a linear relationship between greyscale density and all potential predictors. Multicollinearity undoubtedly occurs within our pool of predictor variables and complicates the selection of a single regression with absolute certainty. As such, we present AICc values for all models (See Supplementary Material) and use the difference from the lowest AICc (ΔAICc) to calculate relative likelihoods following the approach of Burnham et al. (2011). We formally discuss models whose likelihoods are >0.5, which is equivalent to a ΔAICc of 1.4, and can be interpreted as the optimal model being at most twice as likely as an alternate model. Furthermore, we provide comprehensive correlation matrices for all variables in each scenario to evaluate the degree to which our stepwise regression selects unique predictors (See Supplementary Material).

Finally, we give special attention to the relationship between greyscale density and exoskeleton dissolution. In a prior study, a subset of the same field-collected samples analyzed here showed evidence for dissolution that likely occurred after the time of z5-megalope molt, but before field collection (Bednaršek et al., 2020). As mentioned above, more severe dissolution was found at sampling locations with steeper vertical gradients in calcite saturation state (ΔΩ_c-60), and was suggested to be a short-term OA coping mechanism. Given that dissolution has been shown to decrease μCT -based density estimates in other marine calcifyers (Iwasaki et al., 2019), it would not be unreasonable to expect a strong correlation between either the semi-quantitative dissolution estimates of Bednaršek et al. (2020) or ΔΩ_c-60 and our greyscale density data.

Given multicollinearity among potential environmental predictors, we also attempted to isolate the influence of pCO₂ on megalope exoskeleton density through a controlled experiment with pea crabs (Pinnixa sp.). While Dungeness megalopae would have been preferable for this test, attempts to culture the species from the egg to the megalope stage were unsuccessful, at which point wild-caught zoeae were not present in the regional environment. Given that pea crabs and Dungeness crabs are both Eubrachyuran crustaceans and that the larvae can occur in similar habitats, it seems plausible that their biomineralization strategies might also be similar and these experiments still provide information on whether a high pCO₂ environment affects the initial exoskeleton density of megalopae.

We μCT scanned a series pea crabs reared from zoeae to megalopae at ambient (400 μatm) and elevated (2800 μatm) pCO₂. The zoeae were collected in light traps by Swinomish Tribal biologists at Coronet Bay in Puget Sound (48.398033, -122.628037) on June 3, 2019, then reared individually in well plates inside CO₂-chambers at the NOAA Mukilteo Research Station. The CO₂ chambers and crab well plate culture methods are described in McElhany et al. (2022). Initially, 192 zoeae were partitioned equally into four CO₂ chambers (two replicate chambers per treatment). Chambers were maintained at 10°C with a 12 hour on/off light cycle. Zoeae were regularly fed Artermia salina at a target concentration of 1 nauplius per milliliter every 3 days. Those zoeae that survived and molted to the megalope stage were maintained in culture for 14 days, which we considered to be long enough for them to experience any potential dissolution, but before they molted to juveniles. Approximately 14 days after molting, megalopae were preserved in 100% ethanol. A subset of the megalopae were used for μCT analysis (400 μatm, n = 8; 2800 μatm n = 9). We note that the optimal instrument parameters for Pinnixa sp. specimens were subtly different (55kV, 160 mA) than those for M. magister specimens so raw greyscale values are not directly comparable between the two species. However, because analyses were internally consistent for a given species, relative differences in integrated greyscale could still detect the direction of any density change.

μCT greyscale values for field-collected megalopae are summarized in Table 1 . Resolvable differences among the nine sites varied from a minimum of 13.93 at station 127 to a maximum of 16.83 at station 100. There was no obvious relationship between greyscale and latitude or distance from shore.

The best relationships between measured greyscale and environmental variables were always univariate, and are summarized in Table 3 . Both int30 and int60 scenarios yielded modestly significant models (AICc = 26.70, r ² = 0.5, p = 0.03), in which maximum DO was identified as the optimal predictor. Because maximum DO occurred shallower than 30 m, the regression equations and statistics are identical for both assumed habitat depths. DVM30_xx, molt30₁₀ and molt30₂₀ scenarios all produced weak models (AICc = 27.90-28.89, r ² = 0.17-0.28, p = 0.22-0.36) with inconsistent predictors. DVM60_xx, molt60₁₀ and molt60₂₀ scenarios were stronger, but still only marginally significant (AIC = 23.33-25.28, r ² = 0.47-0.60, p = 0.04-0.09). A molt30₃₀ model based on minimum pH at the time of molt was the strongest of any considered (AIC = 19.98, r ² = 0.77, p = 0.01; Figure 5A ), but a second model from this scenario based on minimum T at the time of molt performed nearly as well (AIC = 20.04, r ² = 0.77, p = 0.01; Figure 5B ). A molt60₃₀ model with T at the time of molt as an optimal predictor also performed well (AICc = 20.70, r ² = 0.74, p = 0.01; Figure 5C ).

Combining all possible predictors into a single pool in all30_xx and all60_xx scenarios selects minimum in-situ pCO₂ at the site of collection as the optimal predictor (AICc = 22.27, r ² = 0.68, p = 0.02) for molt times of 10 and 20 days ( Table 3 ). Because minimum in-situ pCO₂ occurs shallower than 30 m, both assumed depth habitats yield the same result. However, in cases where molt was assumed to occur 30 days prior to collection (i.e. all30₃₀ and all60₃₀ scenarios), results were identical to molt30₃₀ and molt60₃₀ scenarios, respectively. That is, the all30₃₀ scenario selected minimum pH at the time of molt, and the all60₃₀ scenario selected T at the time of molt. The AICc for the all30₃₀ model based on minimum T at the time of molt is only 0.06 higher than that for the optimal model based on minimum pH at the time of molt, suggesting it is only 3% less likely and worthy of consideration. In summary, minimum pH at the time of molt assuming vertical migrations over the upper 30 m is the strongest overall predictor of greyscale density, but two alternate models based on minimum T at the time of molt (assuming migrations over the upper 30 m) and average T at the time of molt (assuming migrations over the upper 60 m) are also strong.

We find an insignificant correlation between exoskeleton density and semi-quantitative dissolution (r² = 0.27, p = 0.23), and only a marginal correlation with ΔΩ_c-60 (r² = 0.50, p = 0.03). Consistent with this, the all60₃₀ scenario model based on ΔΩ_c-60 has a ΔAICc of 3.70, suggesting that the model based on T at the time of molt is about 6 times more likely than one based on ΔΩ_c-60. Similarly, the all30₃₀ scenario based on ΔΩ_c-30 has a ΔAICc of 6.02, implying that the model based on minimum pH at the time of molt is about 20 times more likely than one based on ΔΩ_c-30.

The Pinnixa sp. crab were reared under controlled CO₂ conditions for an average of 39 days (min = 23, max = 57), with an average time as zoeae of 26 days (min = 11, max = 46) and the average time as megalopae prior to sampling of 14 days (min = 11, max = 21). There was no statistical difference between the pCO₂ treatments in the duration of zoea or megalope stage. However, three of the crabs in the low-pCO₂ treatment went through a zoeal molt (z4 to z5) then the molt to megalope (z5 to megalope), whereas all other crabs entered the experiment at the z5 stage and experienced only the z5 to megalope molt. The temperature in all chambers was constant with an overall mean of 10.0°C (s.d. = 0.2°C). Spectrophotometric pH values were not statically different among the replicate chambers, with mean low-pCO₂ treatment pH of 8.04 (s.d. = 0.05) and the mean high-pCO₂ treatment pH of 7.24 (s.d. = 0.03). Ω_c calculated with seacarb.R (Gattuso et al., 2021) using pH and alkalinity input parameters was 2.55 (s.d. = 0.24) and 0.45 (s.d. = 0.03) for the low and high pCO₂ treatments, respectively, with alkalinity based on site-specific salinity relationship described in Trigg et al. (2019). The greyscale value of 22.4 for pea crab megalopae reared from zoeae at the 2800 matm pCO₂ treatment was effectively identical to the 22.3 measured in specimens from the 400 matm treatment, implying no significant change in exoskeleton density between the two pCO₂ treatments.

Our evaluation of environmental controls on exoskeleton density suggests that conditions at the time of z5-megalope molt, 30 days prior to collection may be especially important. In particular, minimum pH at the time of initial exoskeleton formation emerges as the single best predictor of exoskeleton density, and suggests that density increases by about 16 greyscale units per pH unit. The highest densities are associated with pH values that never fall below about 7.58 during the 24 hours assigned to z5-megalope molt using the DVM30 behavior, while stations where megalopae had lower-density exoskeletons (e.g. 94, 106, 114, and 120) typically exhibit minimum pH values of ~7.5. Hindcast trajectories suggest that low minimum-pH environments typically occur nearshore the western coast of Vancouver Island, but that there is no unique location for relatively high minimum pH values. For example, Figure 3 suggests that megalopae with denser exoskeletons collected at station 100 were more likely to molt from zoeae further offshore western Vancouver Island, while those at station 115 apparently molted mostly in the Salish Sea.

To our knowledge, no study has specifically investigated how environmental factors influence Dungeness crab megalope exoskeleton formation during the molt from zoea, but considerable work supports the plausibility of minimum pH being an important variable. Studies of larval crab survival and development at variable pH values often find higher mortality and slower progression through molt stages at lower pH or higher pCO₂ conditions (Descôteaux, 2014; Schiffer et al., 2014; Miller et al., 2016; Giltz and Taylor, 2017; Gravinese et al., 2018; Reinhardt, 2020). Minimum values may be particularly important, and duration-dependent pH thresholds of 7.4 to 7.52 have been identified for larval decapods (Bednaršek et al., 2021), while a pH threshold of 7.43 has been suggested for the survival of blue crab zoea exposed to diel cycling (Tomasetti et al., 2021). A similar threshold response has been observed in juvenile blue king crab (Paralithodes platypus), which showed no change in growth or mortality at pH values of 7.8 and 8.1, but exhibited increased mortality and decreasesd growth at pH = 7.5 (Long et al., 2017). While we cannot be certain that exoskeleton density exhibits a threshold response to minimum pH, our lowest measured greyscale value of ~15 corresponds to a minimum pH at the time of z5-megalope molt of ~7.5 ( Figure 5A ) that is similar to previously identified thresholds, making it a plausible scenario. Regardless, it appears that the exoskeleton density of Dungeness megalopae are particularly sensitive to short-lived low pH extremes during the z5-megalope molt, and we suggest that greyscale density provides a metric for tracking the severity of those impacts.

The apparent link between minimum pH and exoskeleton density could reflect additional energetic costs at lower pH that may compromise a megalope’s ability to precipitate its exoskeleton. Assuming studies of adult crab are applicable, the molt from z5 to megalope can be broadly divided into a pre-molt stage that includes the preparation for and actual shedding of the old cuticle, and a post-molt stage that consists of new exoskeleton mineralization. The pre-molt stage is characterized by active uptake of monovalent cations (e.g. Na⁺, K⁺), which drive an osmotic flux of water into the crab, resulting in large weight gains (Cameron, 1989). Calcium is passively transported across the gills and is stored to be used later during mineralization (Wheatly, 1999; Luquet and Marin, 2004; Luquet, 2012). Immediately after molting, the activity of Ca-ATPase enzymatic pumps and bicarbonate transporter proteins increase to actively bring calcium and bicarbonate into the cuticle (Cameron, 1989; Abehsera et al., 2021). Simultaneous active removal of protons from the cuticle promotes the conversion of bicarbonate to carbonate ion, thereby boosting cuticle Ω_c and facilitating rapid calcite precipitation into a complex chitin-protein matrix (Cameron, 1985, 1989; Cameron and Wood, 1985; Luquet, 2012; Figure 6 ). Protons produced as a calcification byproduct must also be actively pumped away into the ambient seawater to prevent hemolymph acidosis (Cameron, 1985, 1989; Cameron and Wood, 1985; Henry et al., 2012).

The numerous instances of active transport involved in molting and reprecipitation of new exoskeleton ( Figure 6 ) indicate it is an energy intensive process, and the energy required to reach a given cuticle Ω_c would increases at lower ambient seawater pH values. For example, the presence of additional protons in ambient seawater at lower pH would mean more protons would need to be actively removed from the site of calcification to boost Ω_c to the same degree of supersaturation, which would require additional energy. If this extra energy was not available, cuticle Ω_c would decrease and may compromise exoskeleton formation. Considerable evidence suggests that coping with low pH conditions could require megalopae to divert energy to other functions thereby leaving less energy for active pumping and calcification and further compromising the ability to reach a given Ω_c. While exposure to low pH or elevated pCO₂ conditions has been shown to increase metabolism, oxygen consumption and ATP production in larval crabs (Schiffer et al., 2013; Trigg et al., 2019), this additional energy may disproportionately go to immune function (Niemisto et al., 2021). Similarly, juvenile isopods have been shown to reduce their carbonic anhydrase concentrations at high pCO₂ conditions (Turner et al., 2016), which would decrease the efficiency of proton removal (Henry, 1984; Cameron and Wood, 1985) supporting the idea that exoskeleton formation is not a priority for energy expenditure at lower pH values.

When viewed collectively, a possible mechanism linking low pH to reduced exoskeleton density might be as follows: Energy limitations reduce the transport of calcium and bicarbonate to the site of exoskeleton precipitation, and reduce the removal of protons from it. Both contribute to a lower cuticle Ω_c, which has been shown to decrease skeletal density and compromise crystal structures in other marine calcifiers (Cohen et al., 2009; Mollica et al., 2018). While additional work is required to confirm this specific hypothesis, it seems likely that an imbalance between the energy needed and the energy available for exoskeleton formation plays a prominent role in explaining the association between lower density exoskeleton and minimum pH at the time of z5-megalope molt.

Considerable evidence suggests that the net response of larval crabs to OA can vary due to interactions among other variables such as temperature, DO and food availability (e.g. Small et al., 2015; Gravinese et al., 2018; Trigg et al., 2019; Niemisto et al., 2021; Tomasetti et al., 2021). This is not surprising given that metabolism, respiration and other physiological processes are linked to energy balance, and thus the ability to cope with low pH conditions. While available evidence supports our selection of minimum pH at the time of z5-megalope molt as the most plausible single predictor of megalope exoskeleton density, models based on mean and minimum T at the time of z5-megalope molt also performed well and we further evaluate the importance of these variables.

Dungeness zoea appear to be adapted to temperatures between approximately 10-14°C, outside of which thermal stress can impact growth and development (Reed, 1969; Sulkin and McKeen, 1994, 1989), increase metabolism and carbohydrate consumption, and deplete lipid concentrations (Brown and Terwilliger, 1999; García-Guerrero et al., 2003; Green et al., 2014; Wittmann et al., 2018; Torres and Giménez, 2020). While most studies explore hyperthermal stress, hypothermal stress is also possible, and would generally leave less energy for exoskeleton formation, thereby potentially impacting exoskelton density in a way similar to the pH mechanism described above. Our hindcast experiments indicate that all field-collected megalopae likely molted at mean temperatures below the optimal 10-14°C range, and experienced diel minima of 4.1 to 5.8°C, potentially making them more sensitive to low pH conditions. In light of this, it may be premature to attribute changes in megalope exoskeleton density exclusively to minimum pH at the time of z5-megalope molt, and minimum temperature may also be important.

The suggestion that M. magister megalopae are sensitive to minimum pH is somewhat at odds with our Pinnixa sp. culture experiments, which show no change in exoskeleton density between mean pH values of 7.24 and 8.04. While inter-species differences are a simple explanation for this apparent discrepancy, they are unsatisfying given the broad similarities between larval M. magister and Pinnixa sp. described above. Instead, we suggest the energy imbalance proposed to explain exoskeleton density variations among field-collected megalopae was minimized or eliminated by one or more experimental conditions, thereby allowing Pinnixa sp. exoskeleton formation to proceed relatively normally under both treatments.

A potential difference between field and cultured organisms was their access to food. Previous work suggests that unlimited access to prey can effectively eliminate thermal stress-induced mortality among zoeae and megalopae because it fuels compensatory responses that are too energetically costly under food limitation (Torres and Giménez, 2020). Cultured Pinnixa sp. were fed routinely, potentially providing sufficient energy for additional active pumping among zoeae in the high pCO₂ treatment. This may have allowed these individuals to reach the same cuticle Ω_c threshold during molt as the zoeae in the lower pCO₂ treatment, thereby minimizing the influence of pH and allowing an exoskeleton of similar density to be precipitated. While the food available to field-collected M. magister at the time of z5-megalope molt is not known, it is plausible that they experienced greater food limitation and therefore a more significant response to pH.

Early life stage exposure history represents a variation on this idea that could also contribute to the apparent difference between Pinnixa sp. experiments and M. magister observations. In a study of pH carryover effects in Tanner crabs (Chionoecetes bairdi), Long et al. (2016) compared wild-brooded larvae to those derived from ovigerous females conditioned at pH values of 7.5, 7.8 and 8.1. Wild specimens transferred to low pH experimental conditions showed minimal effects, in much the same way as our Pinnixa sp. experiments. In contrast, larvae exposed to low pH since oogenesis had larger impacts on morphology, size, calcification and metabolic rate (Long et al., 2016). Based on this, it is possible that M. magister megalopae with lower density exoskeletons were not only exposed to suboptimal pH at the time of z5-megalope molt, but also prior to this time from oogenesis through zoeal stages. Such prolonged exposure may have depleted energy reserves, leaving little energy for the additional active pumping necessary to cope with low pH environments. On the other hand, if it is assumed that Pinnixa sp. were collected from an environment to which they were well adapted, their exposure to suboptimal pH in the experiments was brief and they may not have experienced such carryover effects. Pinnixa sp. may have entered experiments with energy reserves that facilitated additional active pumping and therefore no change in exoskeleton density. If correct, this possibility highlights the need for longer duration OA experiments that better capture carryover effects and will allow more realistic comparisons with field collected specimens.

Finally, minimum temperature is a noticeable difference between cultured Pinnixa sp. and field-collected M. magister. Mean temperatures inferred at the time of z5-megalope molt for field-collected M. magister varied from 8.6 to 10.2°C, and were similar to the nearly constant 10°C of culture experiments. However, inferred minimum temperatures for field-collected M. magister ranged from 4.1 to 5.8°C and were markedly cooler than any temperature experienced by Pinnixa sp. during the experiment. As discussed above, we cannot rule out the possibility that M. magister are sensitive to minimum temperature, and if this is the case, it may explain why field-collected specimens exhibit an apparent sensitivity to pH that was not observed among cultured Pinnixa sp. This possibility suggests that additional experiments exploring how larval M. magister exoskeleton formation responds to hypothermal conditions during diel cycling could provide important insight into their overall sensitivity to environmental stressors.

The weak correlation between exoskeleton greyscale density and dissolution, and the fact that ΔΩ_c was not selected as a top-performing predictor during our model selection suggests that the two processes are not intimately linked. Instead, we suggest that μCT -based exoskeleton density and SEM-based dissolution severity represent complementary techniques that record responses at different life stages: μCT -based exoskeleton density is set by the conditions at the time of initial megalope exoskeleton formation while SEM-based dissolution severity measures subsequent environmental exposure between the time of exoskeleton formation and collection. Furthermore, the weak relationship between these variables implies either that dissolution does not preferentially remove exoskeleton of a certain density, or that the fraction of exoskeleton lost to dissolution is small, such that the initial density set at the time of z5-megalope molt is not influenced appreciably by subsequent dissolution.

The population-level response of M. magister megalope to environmental variables is ultimately most useful for informed management decisions. However, we cannot immediately assume that the individual megalope response to minimum pH at the time of z5-megalope molt translates to the population level. While constructing a robust relationship between individual- and population-level responses is not possible with our data set, we make a simple, initial evaluation. As discussed above, our lowest measured greyscale value of ~15 equates to a pH of ~7.5, which represents a plausible threshold below which megalopae may not survive to be observed. If this is correct, and the population-level response is similar to that at the individual level, one would expect the regional distribution of megalope populations predicted from this greyscale threshold to resemble observations. To test this possibility, we used exposure history hindcasts for all WCOA 2016 stations within the JSCOPE domain to determine the minimum pH at their inferred z5-megalope molt locations, which were then linearly transformed into greyscale estimates using the optimal model from the all30₃₀ scenario ( Table 3 ). Applying our proposed greyscale threshold value of ~15, we then predicted where megalopae would be present and absent.

Our predicted pattern of megalope distribution compares favorably to the observed distribution from 2009-2017 spring monitoring surveys (Morgan et al., 2019; Figure 7 ). While spatial domains do not perfectly overlap, both show megalopae are more likely to be present at more northerly and offshore sites, but are likely to be absent in more coastal and southern waters where pH at the time of z5 molt may fall below ~7.5. One exception is the Columbia River mouth where our simple threshold model erroneously predicts megalopae to be present. This may reflect additional controls on exoskeleton density or megalope presence-absence near low salinity terrestrial sources such as the Columbia River. Despite this shortcoming, the general similarity over much of the domain suggests that μCT based exoskeleton density has potential to inform population-level megalope responses, and therefore may be a valuable tool for making management decisions.

In addition, we suggest μCT-based exoskeleton density also has the potential to provide more nuanced information on M. magister populations impacted by environmental conditions including OA. For example, field surveys that find megalopae to be present, but with a greyscale density approaching our proposed threshold of 15, could suggest a severely impaired population. Conversely, a location where megalopae are present and have appreciably higher greyscale values could reflect a significantly healthier population. By combining such field data with hindcast trajectories, it may also be possible to identify at-risk habitats thereby supporting more informed management decisions. For example, hindcast trajectories for populations with greyscale values near our threshold could identify locations of z5-megalope molt that are most likely to experience minimum pH values near or below 7.5. These regions could then receive special consideration when designing habitat monitoring programs.