Small hermit crabs Diogenes pugilator and green shore crabs Carcinus maenas were collected in autumn at low tide in the Ria Formosa lagoon (Portugal). Harbour ragworms (Hediste diversicolor) were sourced via a local supplier (Valbaits) from the Setubal lagoon (Portugal). Animals were acclimated for 1 week at pH 8.1 (pH is reported using National Bureau of Standards scale) in large flow-through open-circuit communal tanks (continuous flow of 0.05 L/min) mimicking natural habitats at the Ramalhete Marine Station (CCMAR, Faro, Portugal) in a direct CO₂-controlled system (described in Sordo et al., 2016, 2018; Gregório et al., 2019) with pCO₂ constantly measured and adjusted. Two system seawater conditions were used, either control pH of 8.1 ± 0.015 or pH drop of 7.6 ± 0.008. For a measured total alkalinity of 2,500 μmol/kg SW, the pCO₂ was calculated as 537 and 1,922 μatm in the control and pH drop system water, respectively, according to the CO2calc software (v1.2.0, Robbins et al., 2010). pCO₂ was not estimated in any treatment containing reef buffer, HCl, or dilutions, as total alkalinity was not measured in those, and pCO₂ could therefore have differed from that in the system water (see Supplementary Information for details). Seawater parameters were measured daily at 2.30 pm (mean temperature: 20.16°C ± 1.05°C, mean salinity: 35.67 PSU ± 0.26 PSU, mean dissolved oxygen: 7.55 mg/L ± 0.16 mg/L). A total pool of approx. 100–120 D. pugilator, 150–200 C. maenas, and over 150 H. diversicolor was used during the experimental period, yielding a total of 921 observations. Because no individual marking could be achieved, we aimed to space out the reuse of animals in successive behavioural assays by circulating animals through recovery tanks afterward, before returning them to their stock population tanks once a day. Animal reuse was randomised across treatments and treatments randomised per day to prevent any confounding effects on the measured behaviours. All experiments were conducted in October 2019, except one independent set of observations (n = 40) in C. maenas conducted in September 2018, which was analysed together with the 2019 data with the year treated as a covariate.

A three-way factorial design of pH drop × stress metabolites × donor species was generated to study the effects of acute pH drop, stress metabolites induced by it, and their combination vs. a control containing regular metabolites (control metabolites) (Figure 1B). Metabolite donor species were D. pugilator, C. maenas, H. diversicolor (called conspecific donors), and the potential predator Sparus aurata (called heterospecific donor). Metabolite donors were conditioned for 30 min in seawater at regular pH (pH 8.1, 538 μatm CO₂, putatively inducing control metabolite release), or pH drop (pH 7.6, 1,922 μatm CO₂, putatively inducing stress metabolite release). Recipient species were D. pugilator, C. maenas, and H. diversicolor which received water containing either conspecific or heterospecific metabolites. To achieve the full factorial design, recipient species received either control or stress metabolites and were tested in either control pH (8.1) or pH drop (7.6), by re-adjusting the pH of the conditioned water before each bioassay (with the pH verified with an accuracy of 0.02 units using a portable pH metre). Overall, four different experimental treatments were obtained: (i) putative control metabolites at pH 8.1 (CM), (ii) putative control metabolites at pH 7.6 (pH drop), (iii) putative stress metabolites at pH 8.1 (SM), (iv) and putative stress metabolites at pH 7.6 (pH drop + SM). Testing conspecific and heterospecific metabolites with three recipient species amounted to a total of 12 treatments. Prior to use in recipient species, conditioned seawater from H. diversicolor and S. aurata (0.0067 fish/L) was 10-fold diluted in fresh system seawater at the desired pH. To explore the possibility that stress metabolites are equivalent to highly concentrated control metabolites, we additionally tested the effects of undiluted S. aurata control metabolites at pH 8.1 in H. diversicolor and D. pugilator. See Supplementary Information for further details on treatments (Supplementary Table 1) and water conditioning (Supplementary Methods).

For each of the four conditions, we measured the time to find a feeding cue (1/10 diluted mussel juice) vs. a mock (seawater) cue in D. pugilator and C. maenas, or to bury the head entirely in the sand in H. diversicolor, and dubbed this variable “time-to-success.” Behavioural assays were terminated once the animal successfully grabbed the ballasted sponge containing the feeding cue with their pincers (D. pugilator and C. maenas) or buried its entire head (H. diversicolor), or at a maximum time of 300 s. Both feeding and burrowing behaviours are tested response variables in crabs (de la Haye et al., 2012; Wang et al., 2018) and ragworm (Bhuiyan et al., 2021) exposed to pH drop. Additionally, avoidance behaviour was binary coded. In D. pugilator and C. maenas, these avoidance behaviours included freezing (suddenly retreating into the shell or attempting to burrow, sudden and lasting arrest in locomotion for at least 120 consecutive seconds or more than 150 s in total), and escaping (walking along the walls or retreating into a corner of the tank for at least 120 consecutive seconds or more than 150 s in total). Such freeze and escape behaviours indicate danger in crustaceans (Katz and Rittschof, 1993; Perrot-Minnot et al., 2017; Tomsic et al., 2017). In H. diversicolor, avoidance behaviours consisted of freezing, which might be accompanied by spread of jaws, sideways-undulating behaviour, and formation of a noticeable slime cap whilst outside the sediment. These behaviours indicate extreme stress in marine polychaetes (Mouneyrac et al., 2003; Burlinson and Lawrence, 2007; McBriarty et al., 2018). The feeding cues were randomly assigned left or right to the crabs. See Supplementary Methods for further details on behavioural assays. All data is given as Supplementary Dataset.

Avoidance behaviours were analysed using generalised linear models for logit regression with binomial distribution using the stats R package (R Core Team, 2020). Time-to-success was modelled for the time to reach a feeding cue in D. pugilator and C. maenas or to burrow the head in H. diversicolor depending on the three predictors: pH drop, stress metabolites, and donor. Time-to-success data was analysed with a time-to-event analysis (also called survival analysis) using Cox proportional hazard models from the survival R package (Therneau and Grambsch, 2000; Therneau, 2021). The event being the success of food cue location (in crabs) or burrowing the head in sediment (in worms), the analysis was hereafter referred to as “time-to-success analysis.” The exponentiated estimates (hazard ratios) from the Cox proportional hazard models were expressed as “success ratio.” Success ratios were visualised using the plot_model function from the sjPlot R package (Lüdecke, 2021). The “success probability” (the probability of the feeding or burrowing event occurring at any time over 300 s) over time was represented by Kaplan-Meier curves drawn using the survminer R package (Kassambara et al., 2020). Animals not reaching the feeding cue or not burying their head were censored and assigned the maximum time of “300 +”.

All statistical models always included the main effects and interaction terms of the binary predictors: pH drop (pH = 8.1 vs. pH = 7.6), stress metabolites (stress metabolites vs. control metabolites), and donor (conspecific vs. heterospecific). Covariates (year, number of water uses, crab size, where relevant; see section 2.2 in Supplementary Information) were excluded from models when deemed insignificant compared to the null model by at least two model selection methods among the following: an ANOVA Chi-square test from the stats R package (R Core Team, 2020), the Akaike Information Criterion (ΔAIC), and Bayesian Information Criterion (ΔBIC) from the AICcmodavg R package (Mazerolle, 2020). Overall model fit p-values were retrieved from the ANOVA Chi-squared test or the Likelihood ratio test for generalised linear models and Cox-proportional models, respectively. Pairwise comparisons between treatments involved in the interaction term between the three predictors were obtained from statistical models using the emmeans R package (Lenth, 2021), wherein false discovery rate p-value adjustments were applied for post hoc term-wide multiple testing.

To investigate the possibility that stress metabolites are equivalent to highly concentrated control metabolites (which were present in CM and pH drop treatments but in 10-fold dilution), additional tests compared the behavioural response of D. pugilator and H. diversicolor to 10-fold diluted stress metabolites and control metabolites to that of undiluted control metabolites from S. aurata (see Supplementary Information, section 2.4). The full dataset may include animals tested at most twice over the course of the experimental period. We therefore performed random bootstrapping of 50% of the dataset (Supplementary Information, section 2.5). All statistical analyses were conducted in RStudio (R Studio Team, 2020) with a significance threshold of P ≤ 0.05. See R script and Supplementary Information for further details.

The time-to-success analysis was not conclusive for the stress metabolite predictor in D. pugilator (Figure 2A and Table 1, Z = −0.66, P = 0.5073). On the other hand, the Cox proportional hazard model (overall Likelihood ratio test model fit: P = 0.06) found that both pH drop (feeding success ratio = 0.47, Z = −2.14, P = 0.0325) and metabolites from S. aurata (feeding success ratio = 0.35, Z = −2.76, P = 0.0058) had significant negative effects on the feeding success ratio of D. pugilator. In addition, there was a significant interaction between pH drop and metabolite donor terms on the feeding success of D. pugilator (Z = 2.19, P = 0.0289). Overall, pH drop induced a significantly lower success (risk) score in the conspecific treatment (Figure 2B). Post hoc analyses evidenced that all treatments (CM, pH drop, SM, and pH drop + SM) induced similar feeding time responses in both the conspecific and heterospecific donor groups (Supplementary Table 2).

Avoidance responses of D. pugilator did not depend on pH drop (Z = 0.25, P = 0.8049) nor on stress metabolites (Z = −0.25, P = 0.7995, overall Chi-square test model fit: P < 0.0001, Table 2), but were significantly more pronounced when metabolites originated from S. aurata (69%) instead of conspecifics (26%, Z = 2.46, P = 0.0139, Figure 3 and Table 2). After splitting the display of avoidance behaviour by donor, pairwise comparisons failed to find differences between the four treatments (CM, pH drop, SM, and pH drop + SM), but confirmed that the donor effect existed across all four treatments (Supplementary Table 3 and Figure 3). However, partitioning the avoidance behaviour of D. pugilator into freezing and escaping responses evidenced that escaping significantly increased with S. aurata control metabolites only when tested in pH drop (Supplementary Figures 1, 2 and Supplementary Tables 4–7, see Supplementary Results).

The Cox proportional hazard model (overall Likelihood ratio test model fit: P = 0.009) showed that the feeding success ratio of C. maenas did not significantly vary with pH (Z = −0.80 P = 0.4247), metabolites (Z = 0.68, P = 0.4989), nor donor (Z = 1.09, P = 0.2773) terms (Figure 2 and Table 1). Nevertheless, the interaction terms were significant. Post hoc tests revealed that time-to-success responses were similar across treatments CM, pH drop, SM, and pH drop + SM when C. maenas received conspecific metabolites (Supplementary Table 8). This pattern changed when S. aurata was the metabolite donor as evidenced by significantly lower feeding success scores in C. maenas exposed to pH drop (Z = 2.97, P = 0.0179), SM (Z = 2.48, P = 0.0394), and the trend for pH drop + SM (Z = 2.20, P = 0.0563), compared to the control CM.

In C. maenas, the predictors pH drop (Z = −1.23, P = 0.2195), stress metabolites (Z = −1.23, P = 0.2195), and donor (Z = −0.56, P = 0.5785) did not alter avoidance responses but their interaction terms were significant (overall Chi-square test model fit: P < 0.0001, Table 2 and Figure 3). Pairwise comparisons showed no differences between the four treatments (CM, pH drop, SM, pH drop + SM) in avoidance patterns of C. maenas receiving conspecific metabolites. Conversely, C. maenas facing S. aurata metabolites while experiencing treatments of pH drop (94%, Z = −3.60, P = 0.0020), SM (75%, Z = −3.28, P = 0.0031), and pH drop + SM (56%, Z = −2.33, P = 0.04) significantly increased their avoidance display compared to control CM (18%). Moreover, pH drop combined with stress metabolites (pH drop + SM treatment) instead of control metabolites (pH drop treatment) significantly lowered avoidance behaviours (Z = 2.21, P = 0.0405, Supplementary Table 9 and Figure 3). Partitioning the avoidance behaviour of C. maenas into freezing and escaping responses evidenced that escaping explained the avoidance response whereas no clear patterns were observed in the freezing behaviour (Supplementary Figures 1, 2 and Supplementary Tables 10–13, see Supplementary Results).

Burrowing success responses of H. diversicolor did not depend on the donor species (P = 0.1093, Figure 2 and Table 1). However, the Cox proportional hazard model (overall Likelihood ratio test model fit: P < 0.0001) evidenced that both pH drop (burrowing success ratio = 0.30, Z = −3.91, P < 0.0001) and stress metabolites (burrowing success ratio = 0.35, Z = −3.42, P = 0.0006) terms significantly altered burrowing time responses of H. diversicolor. Significant interactions between the main predictors required to decipher the involvement of within-donor treatment effects (Supplementary Table 14). Post hoc tests revealed that pH drop (Z = 3.91, P = 0.0006), SM (Z = 3.43, P = 0.0012), and pH drop + SM (Z = 3.47, P = 0.0012) treatments significantly lowered burrowing success scores compared to the control CM in H. diversicolor exposed to conspecific metabolites. Moreover, H. diversicolor facing S. aurata metabolites had lower burrowing success scores in the pH drop + SM treatment compared to the control CM (Z = 3.35, P = 0.0049).

H. diversicolor displayed more avoidance in response to S. aurata metabolites (47%) compared to conspecific metabolites (20%, Z = 1.09, P = 0.0006, overall Chi-square test model fit: P < 0.0001, Table 2 and Figure 3). H. diversicolor exposed to S. aurata metabolites were not affected by pH drop (Z = −0.01, P = 0.9906) whereas stress metabolites induced significantly more avoidance than control metabolites (Z = 4.31, P < 0.0001, overall Chi-square test model fit: P < 0.0001, Table 2). There were no differences across pH drop, SM, and pH drop + SM treatments in H. diversicolor receiving conspecific metabolites (Supplementary Table 15). Pairwise comparisons revealed that the SM treatment (93%) triggered significantly more avoidance than CM (33%, Z = −4.31, P = 0.0001) and pH drop + SM treatments (51%, Z = 3.92, P = 0.0003). Specimens of H. diversicolor exposed to S. aurata metabolites displayed different avoidance responses in the control treatment CM (Supplementary Video 1) compared to pH drop (slowed-down burrowing response, Supplementary Video 2), SM (sideways-undulating behaviour with body flips, Supplementary Video 3), and pH drop + SM (raised head and spread mouth parts with freezing behaviour in Supplementary Video 4 and slowed-down movement with sideways undulating behaviour in Supplementary Video 5).

Importantly, the behavioural responses of H. diversicolor and D. pugilator remained unchanged even when we used undiluted control metabolites (see Supplementary Information, section 2.4, Supplementary Figure 3 and Supplementary Tables 16, 17). Of note, the avoidance response of H. diversicolor and D. pugilator faded as water was reused (Supplementary Information, section 2.2, Supplementary Figure 4).

Furthermore, random bootstrapping of 50% of the dataset (Supplementary Information, section 2.5, Supplementary Figure 5 and Supplementary Table 20) did not change the outcome of time-to-event analysis for any species.

In this study, we performed over 900 behavioural observations, and our results show species-specific responses, donor-dependent effects, as well as sometimes large within-treatment variability. Sparus aurata-conditioned water, whether it contained control metabolites or pH stress-induced metabolites, elicited avoidance in all recipient species which can be attributed to antipredator behaviours. Despite the donor effect, however, much of the behavioural variance was explained by the effects of pH drop and stress metabolites, their interaction, or their interaction with the donor term (Figure 2).

We found that pH drop lowered success ratios in feeding and burrowing in Diogenes pugilator and Hediste diversicolor, respectively. Such low pH-induced impairments of fitness-relevant behaviours were previously found in hermit crabs Pagurus bernhardus and P. tanneri as well as in the Japanese stone crab Charybdis japonica and may indicate impaired decision-making (de la Haye et al., 2011, 2012; Kim et al., 2015; Wu et al., 2017). Past literature also reported delayed burrowing responses in the harbour ragworm H. diversicolor exposed for 28 days to pH drop (Bhuiyan et al., 2021), and an inverse relationship between burrowing and pH in the king ragworm Alitta virens (Batten and Bamber, 1996). These behavioural alterations may reflect the cost of acid-base regulation to counteract hypercapnia (Pörtner, 2008), as marine ragworms overexpress the acid-base regulator Carbonic Anhydrase (CA) at the expense of energy reserves in response to acidification (Freitas et al., 2016; Wage et al., 2016). In the longer term, behavioural costs of an altered burrowing activity may extend to crucial ecological functions as shown by the diminished bioturbation activity of H. diversicolor in similar conditions (Bond, 2018). In contrast, Carcinus maenas did not delay its feeding response to sudden pH drop in presence of conspecific metabolites, which is in line with meta-analyses showing that some aquatic arthropods are more robust to ocean acidification conditions (Kroeker et al., 2013; Wittmann and Pörtner, 2013; Clements and Darrow, 2018). The resilience to pH stress in C. maenas may be due to their pH regulatory capacities allowing them to maintain feeding during short-term assays, but they may show altered responses in longer-term assays—for instance after 10 weeks—as their energetic requirements increase (Appelhans et al., 2012; Clements and Darrow, 2018). Alternatively, the fact that C. maenas did not respond to pH drop may arise from differences in the type of cues that are presented or that not all behaviours are similarly affected by pH drop (Richardson et al., 2021). Short-term hypercapnia does not seem to be a strong stressor in C. maenas, due to its pH-variable intertidal habitat (Fehsenfeld et al., 2011). Other than reduced success ratios, pH drop itself did not induce any avoidance behaviours indicative of stress.

Besides the direct effect of pH drop on behaviour, previous studies on ocean acidification also found contrasting effects of pH drop on predator-prey interactions (Draper and Weissburg, 2019). Here, we observed an increased escaping tendency of C. maenas in response to S. aurata control metabolites in pH drop but not in control pH. Although this was not the main focus of our study, it could mean that pH drop renders predator odour more potent for its prey, for example through pH-dependent changes in odour molecular structure, receptor binding, or information processing (Munday et al., 2009; Roggatz et al., 2016, 2019; Schirrmacher et al., 2020). Our results contrast with earlier reports of decreased predator avoidance (e.g., in snails; see Jellison et al., 2016; Froehlich and Lord, 2020). Overall, our data suggests that a short-term acute pH drop has the ability to impair decision-making and predator-prey interactions, and in turn alter fitness-relevant behaviours. Nevertheless, future studies should aim to disentangle the effects of pH drop from that of pCO₂, which ideally also would test short, medium, and long-term responses.

Despite the above-mentioned species-specific direct effects of pH drop on behaviour success ratios, we could show here that pH drop also has indirect effects on behaviour by altering chemical communication. We found that stress metabolites released by animals exposed to pH drop altered feeding and burrowing, and increased avoidance behaviours, but that the response was donor and recipient species-specific. Hermit crabs D. pugilator did not react to any stress metabolites. Significant responses in the other two recipient species, however, were directionally equivalent to those induced by pH drop itself since it took longer to find a food cue or to burrow the head in the presence of stress metabolites. Although behavioural stress response propagation through chemical communication is well documented for alarm substances and disturbance cues upon predation stress (Giacomini et al., 2015; Abreu et al., 2016; Mathuru, 2016), our results show here that chemical communication induced by abiotic stressors such as pH drop can trigger the release of waterborne chemicals that in turn trigger a behavioural stress response in its recipients. Stress metabolites conditioned by the potential predator S. aurata induced more avoidance behaviours and delayed feeding in C. maenas and burrowing in H. diversicolor, compared to S. aurata control metabolites. Similarly, the presence of conspecific stress metabolites significantly lowered H. diversicolor burrowing success ratio compared to conspecific control metabolites.

On the other hand, C. maenas did not react to conspecific stress metabolites. The different behavioural effect of conspecific stress metabolites between crabs and H. diversicolor has several possible explanations ranging from different defence strategies related to stress risk perception and evaluation (Hazlett, 1985; Bairos-Novak et al., 2017, 2018; Goldman et al., 2020b) to different types, concentrations, or ratios of the released chemicals that serve as risk cues (Júnior et al., 2010; Morishita and Barreto, 2011). As we have shown here a pH drop to 7.6 is not a strong stressor in crabs (Fehsenfeld et al., 2011) which may explain that we failed to observe responses to water from pH drop-conditioned conspecific green shore crabs. However, the presence of stress metabolites conditioned by S. aurata caused a marked drop in the success ratio of C. maenas at low (but not control) pH, evidenced by significant interaction terms.

Whilst we did not investigate their chemical structure, our experiment allows us to characterise novel properties of stress metabolites. Stress metabolites, like disturbance cues, potentially consist of regularly excreted metabolites such as urea and ammonia (Bairos-Novak et al., 2017; Shrivastava et al., 2019). However, the fact that behavioural responses of H. diversicolor and D. pugilator remained unchanged even when we used undiluted control metabolites might mean that stress metabolites are not just up-concentrated control metabolites (at least in the tested range of 10-fold control metabolites), or that the cocktail of chemicals contains much higher ratios of stress cues relative to control cues. We also observed that avoidance behaviours by both C. maenas and H. diversicolor became somewhat less pronounced with subsequent water uses, suggesting that the recipients may be able to discriminate degraded vs. fresh metabolites and react accordingly to the degree of threat they may indicate (Fuselier et al., 2009; Bairos-Novak et al., 2018). This shows that stress metabolites induced by pH drop are either not identical with metabolites released in normal pH, and/or are at least more than 10-fold higher concentrated than control metabolites.

The avoidance behaviours we recorded in this study are known indicators of stress. S. aurata stress metabolites elicited atypical freezing, eversion of the proboscis, mucus secretion, flipping and sideways-undulating behaviours in H. diversicolor which were previously described as indicators for a physiological stress response following exposure to copper sulphate (Burlinson and Lawrence, 2007), antiparasitic drugs (McBriarty et al., 2018), and trace metals (Mouneyrac et al., 2003). We also observed increased freeze and escape responses in crabs, which indicate a stress response in crustaceans (Katz and Rittschof, 1993; Perrot-Minnot et al., 2017; Tomsic et al., 2017). The finding that S. aurata stress metabolites induced such responses in potential prey species suggests that responses to disturbance cues are do not depend on the “audience,” similar as what has been observed in tadpoles (Bairos-Novak et al., 2020).

A directionally similar response to pH drop and to the stress metabolites it induces shows that pH drop can have indirect effects through stress propagation (Figure 1A), by which stressed animals directly experiencing environmental stressors negatively influence fitness-relevant behaviours of community members. The stress propagation toward naive conspecific individuals might constitute a “positive stress feedback loop” (Feugere et al., 2021). This was first observed by Hazlett (1985) after freshwater crayfish exposed to water conditioned with heat-stressed conspecifics displayed increased alertness, after returning the conditioned water to normal temperature. Similar indirect effects through chemical signalling from stressed donors toward unstressed, naive recipients exposed to the water conditioned by the directly stressed donors but not to the stressor itself, were also shown in response to different types of biotic stressors such as mock predator chase (Toa et al., 2004; Giacomini et al., 2015), handling (Barcellos et al., 2011), and acute fasting (Abreu et al., 2016), and to physical injury including irradiation (Mothersill et al., 2007) and predation (Frisch, 1938; Mathuru et al., 2012; Oliveira et al., 2013).