Acoustic behaviour of male European lobsters (Homarus gammarus) during agonistic encounters

Sounds can be used by marine organisms to convey information. Numerous studies have demonstrated that marine mammals and fish use sounds to navigate, find food, communicate with conspecifics or even deter predators (e.g. Tyack and Clark, 2000; Ladich, 2015). By comparison, the potential role(s) of sounds amongst marine invertebrates is poorly described (Taylor and Patek, 2010; Edmonds et al., 2016).

For instance, only a few crustacean species have been shown to produce sounds during behavioural interactions. The tropical spiny lobster (Panulirus argus) produces antennal rasps when attacked by predators (Bouwma and Herrnkind, 2009). Mantis shrimp (Hemisquilla californiensis) rumble to maintain their territories against conspecifics (Patek and Caldwell, 2006; Staaterman et al., 2011). Semi-terrestrial crabs (the Ocypodidae) are known to produce stridulations that attract females to their burrows for mating (Popper et al., 2001). Other crustacean species have also been shown to produce sounds, but the lack of relevant behavioural studies does not yet permit validation of potential ecological roles for these sounds. In the temperate coastal waters of Brittany (France), several crustacean species produce a large diversity of sounds, but their ecological roles, if any, are unknown (Jézéquel et al., 2018, 2019).

Lobsters, particularly the American lobster (Homarus americanus), have been identified as a good study model for analysing complex behaviours (Scrivener, 1971; Atema and Voigt, 1995; Huber and Kravitz, 1995). Male H. americanus lobsters exhibit highly aggressive behaviours towards each other. Indeed, they use agonistic encounters to establish and maintain their dominance within a group to gain better access to shelters and females for reproduction (Scrivener, 1971; Atema and Cobb, 1980; Atema and Voigt, 1995). When two individuals meet, they exhibit an array of agonistic behaviours ranging from visual displays to physical contact (Scrivener, 1971; Huber and Kravitz, 1995; Breithaupt and Atema, 2000). The main factor influencing the outcome of an agonistic encounter is body size: larger individuals have a greater chance of winning an encounter (Scrivener, 1971). This results in shorter behavioural sequences compared with those for size-matched lobsters where their aggressive behaviours lead to highly stressful conditions (Atema and Voigt, 1995). The loser of an encounter avoids the winner afterwards, and dominance is maintained through a variety of signals. Chemical signals (i.e. pheromones) released in urine appear to be the main means of preserving the memory of the outcome between pairs of individuals, post-encounter (Breithaupt and Atema, 1993; Karavanich and Atema, 1998; Breithaupt et al., 1999). Recently, Gherardi et al. (2010) and Bruce et al. (2018) showed that visual recognition of specific individuals also plays a role. The ability to recall the outcome of past encounters may help individual lobsters to avoid additional fights and lower their future risk of injury (Breithaupt and Atema, 2000).

A recent study has shown that, similar to H. americanus, the European lobster (Homarus gammarus) also emits buzzing sounds when stressed (Jézéquel et al., 2018). These sounds are produced through the rapid contraction of internal muscles located at the base of their second antennae, which causes the carapace to vibrate (Mendelson, 1969). These ‘buzzing’ sounds are characterized by low frequencies (∼100 Hz) and have a relatively long duration (∼200 ms; Henninger and Watson, 2005; Jézéquel et al., 2018). Ward et al. (2011) suggested that H. americanus may only use these sounds to deter predators. Interestingly, earlier studies indicated that few buzzing sounds were produced during agonistic encounters in male H. americanus and it was then concluded that these sounds do not have a role for intraspecific interactions (Scrivener, 1971; Atema and Voigt, 1995; Atema and Cobb, 1980). Hence, no study has examined the ecological role of these buzzing sounds and only one has described the behavioural patterns in H. gammarus during agonistic encounters (Skog et al., 2009).

The primary aim of the present study was to: (1) test whether male H. gammarus emit buzzing sounds during agonistic encounters, and (2) test the potential role of these sounds as acoustic communication between lobsters. For this purpose, we designed an experimental laboratory set-up whose main feature was a tank containing the lobsters under study. The set-up also combined hydrophones to record their buzzing sounds in the tank, accelerometers on the lobsters to record their carapace vibrations (i.e. the source of the buzzing sounds) and cameras to record animal behaviour. Firstly, we developed a detailed ethogram based on the video recordings of the encounters. Secondly, we analysed the behavioural sequences between dominant and submissive individuals. Lastly, we examined whether the sequences of buzzing sounds produced by two individuals depended on their relative size differences. We then analysed these as call-and-response signals to explore their potential role for the communication of dominance.

All laboratory experiments were carried out at the research facilities of the Océanopolis public aquarium located in Brest (France).

For these experiments, a total of 24 H. gammarus (Linnaeus 1758) male individuals, with carapace length (CL; measured from the eye socket to the posterior carapace margin for lobsters) between 8.7 and 13 cm, were used. They were collected carefully by hand while snorkelling in the Bay of Plougonvelin (Brittany, France) at water depths of between 1 and 10 m. Two samplings were done in May and July 2018; 12 individuals were collected during each session. Only intermoult individuals (following the description in Aiken, 1973) with full sets of undamaged appendages were collected and used for this study.

After capture, lobsters were separated randomly into two groups of 6 individuals each, and then transferred to different holding tanks. One group was held in a large shaded, polyester circular tank (radius 4 m, effective height 1.13 m, seawater volume 14.2 m³). The second group was held in two identical plastic rectangular tanks (1.50 m×1.00 m×0.5 m length×width×effective height; seawater volume 0.75 m³) with 3 individuals per tank, separated by plastic dividers. In the communal tanks, the lobsters' claws were bound with numbered rubber bands to avoid injury. These also identified each individual lobster. All holding tanks were continuously supplied with sand-filtered, UV-sterilized seawater pumped from the Bay of Brest. Temperature, salinity and animal condition were controlled twice a day. During holding, temperature varied between (mean±s.d.) 14.8±1°C (in May and June) and 17.5±0.5°C (in July and August) and salinity between 34.4±0.3 and 34.9±0.1. Animals were fed with fresh pieces of fish (mackerel) and cephalopod (squid) ad libitum. They were kept under the natural photoperiod in the large circular tank, and under a 12 h:12 h photoperiod in the smaller tanks, the daylight condition being simulated by fluorescent light tubes above the tanks. Sections of rigid PVC drainage pipes were provided in abundance as shelters. Animals were acclimatized for at least 1 month in these conditions before they were used in the experiments.

All experiments were done in a dedicated plastic tank (1.13 m×0.73 m×0.5 m; 0.4 m³) placed in a quiet room, isolated from the main activities of the aquarium facilities (Fig. 1). The bottom was covered with a thin layer of sand, 5 cm deep, to provide a foothold for the animals. Two LED light strips (B0187LXUS2, colour temperature 4500 K) were placed 50 cm above the tank to ensure good visibility for video recording by the cameras. The experimental tank was divided into two equal volumes by a removable, opaque, Plexiglas divider (6 mm thick) installed in the middle of the tank prior to introducing the animals (Scrivener, 1971; Huber and Kravitz, 1995; Skog et al., 2009). To do this, plastic gutters were epoxy glued on the vertical sides and along the bottom of the tank. This permitted the divider to easily slide up at the start of each experiment. The edges of these gutters were silicone sealed to eliminate any water exchange while the divider was in place. The barrier prevented the exchange of chemosensory and visual cues between the two lobster opponents before the agonistic encounter was started by removal of the divider.

Sounds were recorded using two pre-amplified hydrophones (HTI-92-WB, High Tech Inc., Long Beach, MS, USA), with a sensitivity of −155 dB re. 1 V µPa⁻¹ and a flat response between 2 Hz and 50 kHz. Hydrophones were connected to a compact autonomous recorder (EA-SDA14, RTSys, Caudan, France) with a gain of 14.7 dB, and were powered by battery to limit electronic self-noise. Recordings were made with a sampling frequency of 156 kHz at 32-bit resolution. Even though buzzing sounds are characterized by low frequencies (∼100 Hz; Jézéquel et al., 2018), we chose a high sampling frequency because we wanted to cover a large frequency band, in case the lobsters produced new sounds during the experiments.

We used two hydrophones in the experimental tank to ensure most of the buzzing sounds emitted by individuals during the agonistic encounters could be recorded. One was placed in each compartment of the tank, 30 cm above the substrate, and they were separated by 55 cm from each other (Fig. 1). Based on our earlier work to determine the acclimation state of the animals used (Jézéquel et al., 2018), this installation did not perturb the individuals.

During preliminary trials, we noted that very few buzzing sounds were recorded by the hydrophones during agonistic encounters. Lobsters emit buzzing sounds through rapid contractions of internal muscles located at the base of their antennae, which vibrate the carapace (Henninger and Watson, 2005); we therefore added accelerometers on their carapaces as a means to detect carapace vibration events, independently of the hydrophones.

One small AX-3 data logger (23×32.5×8.9 mm, mass 11 g; Axivity Ltd, Newcastle Helix, UK) was glued with 3 min underwater epoxy to the dorsal carapace of each lobster, near the eye sockets at the base of the second antennae (Fig. 1). The x-axis was oriented parallel to the longitudinal body axis, which is also parallel to the internal muscles responsible for the carapace vibration (Henninger and Watson, 2005). The accelerometers were set to record acceleration in all three axes (range ±16 g, 156.96 m s⁻²) with a sampling frequency of 3200 Hz and a 13-bit resolution. The accelerometers had a 512 MB memory card onboard. Each accelerometer was waterproofed before attachment by encasing it in a polyethylene film sealed shut with heat-shrink tape. Air trapped inside the polyethylene film made the accelerometer loggers neutrally buoyant in seawater. All accelerometers were marked with unique numbers to associate them with particular individuals. This technique permitted us to link each carapace vibration recorded to an individual and also to validate the buzzing sounds recorded with the hydrophone recordings. As stated above for the hydrophones, we did not observe any evidence that the presence of the sensors perturbed their movements during the experiments.

Visual observations and video recordings were made during all experiments using three GoPro^® HERO3 cameras. Two cameras were placed in the bottom of the tank at either end against the walls, and a third camera was placed 50 cm above the water surface of the tank (Fig. 1). Videos used a recording rate of 29.97 frames s⁻¹ with an image resolution of 1920×1080 pixels.

To ensure that all the data streams could be re-synchronized, we used a synchronization procedure at the end of the experiments. First, the accelerometers were gently taken off the lobsters and placed on the sand in the middle of the tank, and the two lobsters were returned to their holding tanks. Then, five sharp raps were made on the tank walls that could be used to synchronize all three types of recording device (hydrophones, accelerometers and GoPros).

Experiments were performed during June and August 2018 in the experimental tank described above. During each experiment, seawater temperature was measured using a HOBO Pendant G data logger (UA-004-64, Onset Computer Corporation). Seawater temperature in the experimental tank was 17.11±0.14°C (mean±s.d.) in June and 18.44±0.12°C in August.

Agonistic encounters were set up between two categories of lobsters: size-matched male lobster pairs (difference in carapace length, ΔCL<5 mm), and small and large male individuals (ΔCL>5 mm). In fact, larger lobsters are more likely to win a fight if the ΔCL is more than 5 mm between the opponents, but at smaller size differences, the outcome is random (Scrivener, 1971). We formed pairs by taking one individual from each separately acclimated group to ensure that the individuals had no prior knowledge of each other (Karavanich and Atema, 1998). A total of 12 agonistic encounters (6 with ΔCL<5 mm; 6 with ΔCL>5 mm) were set up.

Because communal holding causes a general reduction of aggressiveness in lobsters (Breithaupt and Atema, 2000), we isolated the two selected individuals separately for 24 h in glass-sided rectangular tanks (0.60 m×0.50 m×0.35 m; 0.105 m³) after the accelerometers were attached. This allowed the lobsters to recover from handling. For this step, the bands on their claws were also released. Lobsters were not fed during this period.

The next day, these same individuals were placed in the prepared experimental tank, one on either side of the divider (Fig. 1). Experiments were performed between 16:00 h and 20:00 h. Recordings started when the individuals were placed in the tank. We recorded the first 15 min as control observations of the individuals while they were in isolation in their respective compartments. Next, we lifted the divider and continued recording the agonistic encounters that ensued for another 15 min. This corresponds to the expected minimum time for determining the outcome, according to Scrivener (1971). After the experiment, the accelerometers were removed from both animals, and the lobsters were returned to their holding tanks. Then, the data synchronization procedure (described in ‘Data synchronization’, above) was followed. Afterwards, the experimental tank was drained completely, thoroughly rinsed and refilled with fresh seawater, and the sand was replaced. Each individual was used only once during the study.

Sound files (.wav) from the two hydrophones (30 min recordings each) were archived at the end of each experiment. They were first carefully visualized over the entire frequency band (between 0 and 78 kHz) by using the spectrogram mode in Audacity^® (v2.1.1; www.audacityteam.org) to check for potential biological broadband sounds emitted by lobsters during experiments. Next, sound data were subsampled between 0 and 500 Hz and spectrograms were visualized a second time using custom-written MATLAB scripts (v9.1; The MathWorks, Natick, MA, USA). The characterization of buzzing sounds has been detailed in our earlier study (Jézéquel et al., 2018). As the aim of the experiments in the present study was to understand when these sounds were produced within the behavioural sequences, here we only report the basic descriptive statistics of the buzzing events recorded during the agonistic encounters.

Data from the accelerometers were downloaded using Open Movement GUI software (v1.0.0.37). Accelerometers record movements simultaneously on three axes as the relative change detected in gravitational acceleration, g (1 g=9.81 m s⁻²), and carapace vibrations are known to have the same frequency as their associated buzzing sounds (∼100 Hz; Henninger and Watson, 2005). After examination of the data on all three axes, we observed the strongest signals of the carapace vibrations were recorded on the x-axis, as expected. We thus used only the x-axis data to facilitate their detection among all the other high-amplitude signals related to the lobster movements (e.g. tail flips) by treating the data using a custom-written MATLAB script. We report here the number and timing of carapace vibration events recorded during each agonistic encounter for each individual. We also measured their duration (in ms) and peak frequency (in Hz).

Video analysis consisted of annotating the visible movements performed by each individual during the encounters. Based on the extensive H. americanus literature (see Table S1) and preliminary tests, we built a description of movements (also termed ethogram) by annotating 30 associated movements for five different body parts (antennae, claws, legs, carapace, tail; Table 1). We focused on movements or actions initially instead of ‘behaviours’ because it allowed us to avoid subjective choices related to the sometimes ambiguous behaviours defined in the literature. Movement directions like ‘walking away’ and ‘walking backward’ were identified according to the direction of the body axis relative to the other individual. For example, ‘walking away’ for a lobster was defined as the direction of its rostrum that pointed away from its opponent, but does not necessarily mean it was escaping from its opponent. These 30 movements were annotated for each individual and for all 12 agonistic encounters using the tools in BORIS (v6.3.9; Friard and Gamba, 2016).

Video data from each agonistic encounter comprised video recordings (30 min each) from each of the three cameras used in the experiments. We chose to annotate primarily videos from the plan view camera because this covered the entire experimental area and most of the movements were visible. We completed these observations by analysing the recordings from the two cameras placed in the bottom of the tank. This permitted us to visualize more precisely certain vertical movements made by the lobsters (e.g. high on legs, meral spread). All these annotations were then integrated with the annotation from the plan view camera for subsequent data treatment. Time energetic budgets were made for each movement and each individual (submissive and dominant) as percentages of the total length of the agonistic encounters (15 min).

Statistical analyses were performed using R v3.5.1 (http://www.R-project.org/). The mean percentage total time for each movement was tested for significant differences between dominant and submissive individuals in all 12 encounters. As these data were not distributed normally (Shapiro–Wilk test, P<0.05), the non-parametric Mann–Whitney test (U-test) was used to determine whether their probability distributions were equal. The significance level for null hypothesis rejection was α<0.05. These results permitted us to associate sequences of movements typically shown by dominant and submissive individuals to particular behaviours based on the conventions used in the H. americanus literature (see Table S1).

As our experiments were necessarily brief to avoid injury (15 min long; Scrivener, 1971) and each experiment was also unique, the carapace vibration sequences did not meet the criteria for classical statistical tests (Guarini et al., 2019). Because the development of a behavioural model was beyond the scope of the present work, we only considered whether the sequences of carapace vibrations recorded by the accelerometers on each individual during the agonistic encounters could not have been produced by a random process. Instead of classical tests, we used a definition of randomness for algorithmic complexity that was recently formalized for short series of fewer than 100 characters that are common in behavioural studies (Soler-Toscano et al., 2014; Zenil et al., 2015preprint; Gauvrit et al., 2016).

Algorithmic complexity offers an alternative means to evaluate the existence of ordered patterns in short sequences by assessing the computing effort needed to stimulate them (Zenil et al., 2018). The approach compares a given string with results from randomly selected Turing machines calculating the likelihood that the string could be reproduced by these algorithms. In this definition, a low-complexity string has a higher probability of being generated by a randomly selected Turing machine, and therefore is less likely to have been produced by a random process (see development in Gauvrit et al., 2016). This has the double advantage of producing invariant estimates of complexity for a given observed sequence and that each experiment is treated as unique. In other words, each sequence is only compared with its own realization relative to the Turing machine algorithm. This method does not use thresholds to infer randomness (Zenil, 2015). Instead, it estimates the algorithmic complexity (AC) and an indicator of the computing time required to compress the sequence structure, called the logical depth (LD; Zenil et al., 2018). A longer LD means a non-trivial structure has been found in the sequence.

To apply this method, carapace vibration sequences produced by individuals during the same agonistic encounter were transformed to time-ordered, discrete binary series. Carapace vibrations were assigned to 1, if produced by the dominant individual, or 0 if produced by the submissive individual (e.g. 1000000001010010); the rhythm of the carapace vibrations (i.e. the time between vibrations) and their duration were not represented. This also means we considered that two individuals produced carapace vibrations sequentially (i.e. as ‘call-and-response’) and not simultaneously. Because of the short length of our strings (from 14 to 98 characters), we used the block decomposition method made available through an online tool to access the necessary range of Turing machine states (Soler-Toscano et al., 2014; Zenil et al., 2018; http://complexitycalculator.com/index.html, v3.0). The most conservative settings were used: the largest available maximum block size (12), with no overlap and a two-character alphabet. As the AC and the LD both depend on string length, we report normalized values (as bits per character and steps per character, respectively). Hence, a standardized AC value of 1 or higher would be considered as not differentiable from random. Using a two-character alphabet, when the standardized LD is about 2 or higher, then the process that generated the sequences cannot be distinguished from a random one (Zenil et al., 2018).

Experiments with H. gammarus are not subject to restriction for animal scientific research according to the French legislation and the European Community Council Directive of September 2010 (2010/63/UE). We nonetheless followed the ARRIVE guidelines (Kilkenny et al., 2010) for all the experiments. The animals' health state was checked daily by the authors and the aquariology team of Océanopolis. During experiments, we planned to stop the agonistic encounters between two lobsters before any injury occurred to the animals; this never happened and no lobsters were injured or died during the study. At the end of the experiments, all animals were released back into the area from where they were collected.

When isolated on either side of the divided tank, lobsters wandered freely around the space and did not show any particular movements related to the other individual. When the divider was lifted, the individuals quickly engaged physically in an agonistic encounter (e.g. Fig. 2). Initially, they made a short (<1 min) series of threat displays, typically consisting of: antenna pointing or antenna whipping, claw open, meral spread and high on legs movements. Next, they advanced rapidly with different types of physical claw contact to drive away their opponent. This stage was mainly dominated by claw pushing movements. In 6 of the 12 agonistic encounters (4 with ΔCL>5 mm, 2 with ΔCL<5 mm), the outcome was decided at this stage. In the six other trials, the lobster pairs increased the intensity of the fight by using a variety of claw movements to attack their opponents. These movements, such as claw boxing, claw ripping or claw snapping, were very short in duration and occurred in association with aggressive upward-directed tail flipping. Generally, after these actions, one individual withdrew and assumed the submissive role for the remaining time (Fig. 2).

After this first encounter, which determined the hierarchical status between the two lobsters, each dominant and submissive individual displayed typical groups of movements (Fig. 2, Table 2). Dominant individuals continued to perform physical displays (i.e. meral spread, high on legs and claw open), and often approached the submissive individuals (i.e. walking toward) to re-engage in physical contact (mainly antenna whipping and claw pushing). In contrast, submissive individuals always responded by escaping through physically demanding movements such as walking backward and tail flipping (Fig. 2, Table 2). In particular, submissive individuals used a characteristic submissive posture with the claws extended in front of the animal for much of the period following the first encounter. Finally, when individuals were not making claw contact, the dominant animals were moving actively around the tank such as walking or sand removing, while in contrast, the submissive ones were relatively immobile (i.e. resting) near the tank walls with their claws extended (mean: 44.9% of time; Fig. 2, Table 2).

During the agonistic encounters, we did not record any particular sounds other than the buzzing sounds with the hydrophones. We identified a total of 65 buzzing sounds from 9 of the 24 lobsters tested. In marked contrast, the accelerometer data showed that 23 out of the 24 lobsters tested vibrated their carapace during the agonistic encounters. The only lobster that did not vibrate its carapace was a dominant individual. From these 23 lobsters, a total of 422 carapace vibrations were recorded, meaning that only 15% of the associated buzzing sounds were recorded by the two hydrophones in the tank. Fig. 3 shows an example where two lobsters produced three carapace vibrations during a short period (6 s), and the associated buzzing sounds were only recorded by the closest hydrophone (<20 cm from the animals). However, in most other cases when lobsters vibrated their carapaces, the associated buzzing sounds were not recorded by the two hydrophones at the same time.

We therefore used the number of carapace vibrations as a proxy for the number of buzzing sounds produced by lobsters. No carapace vibrations were detected when lobsters were first separated from each other by the divider. Even though some first encounters were long (up to 3.38 min) with highly aggressive movements between lobsters (e.g. claw ripping), very few carapace vibrations (4.7%) were produced at this time (Fig. 2). In contrast, carapace vibrations were mostly (95.3%) produced after the first encounter (i.e. after hierarchical status was determined), during the stage of repeated approaches by the dominant individuals making threat displays towards the submissive individuals (Fig. 2).

Fig. 4 describes the distribution of all carapace vibrations detected by the accelerometers according to dominant or submissive outcomes. Overall, dominant individuals emitted about half as many carapace vibrations as submissive ones (141 and 281, respectively). Carapace vibrations had durations that varied from about 50 ms to nearly 600 ms, and their peak frequencies varied between 100 and 200 Hz. No carapace vibrations were recorded that began at exactly the same time. These data were also plotted as time series for all 12 encounters (Fig. 5). There are few clear patterns in the series. The total number of carapace vibrations in an experiment between individuals of nearly the same CL (Fig. 5, left) tended to be higher than that in experiments where the ΔCL was >5 mm (Fig. 5, right). Submissive individuals, which were also the only individuals to assume the ‘extended claw’ pose (Table 2), produced carapace vibrations in all encounters and mostly, but not always, while in this pose (Fig. 5). For most agonistic encounters, submissive individuals produced more carapace vibrations than did dominant ones; but in three experiments (Fig. 5C,H,L), the opposite pattern was obtained and the dominant animal vibrated more frequently. In one experiment, the dominant individual was silent (Fig. 5E).

As described above, the carapace vibration series were then expressed as binary, ordered sequences and analysed for their AC and LD. The string standardized values of the AC and LD are given in Fig. 5, in bits per character and steps per character, respectively. The values of both measures (1<AC<3 and 2<LD<4) indicate that the carapace vibration sequences were probably the product of a random process, and by themselves cannot be assimilated to call-and-response type signalling.

This study is the first report of male H. gammarus producing buzzing sounds during agonistic encounters. These sounds were produced by both dominant and submissive individuals during the experiments and were mainly emitted after the end of the first encounter (when claw contact stopped) up until the experiment ended.

Our agonistic encounters resembled descriptions of agonistic encounters published in earlier studies of male H. americanus (Scrivener, 1971; Atema and Voigt, 1995; Huber and Kravitz, 1995) and male H. gammarus (Skog et al., 2009). The initial stage consisted of a threat display between individuals that then quickly engaged in physical claw contacts, which could increase in aggressiveness (e.g. claw boxing) until the withdrawal of one individual (Fig. 2). This losing individual then exhibited submissive behaviours highlighted by a claws extended pose and was less active, while the winner remained active and continued to make approaches and threat displays. At the same time, both individuals produced buzzing sounds.

However, during these experiments, very few buzzing sounds were recorded by the two hydrophones even if they were placed close to the lobsters (<75 cm away). This is consistent with remarks made in previous studies on H. americanus (Scrivener, 1971; Atema and Cobb, 1980; Atema and Voigt, 1995; Ward et al., 2011). For example, Atema and Cobb (1980) stated that ‘the biological significance of such vibrations is unknown; during high intensity fights in aquaria, these sounds were rarely recorded’. Ward et al. (2011) showed, with accelerometry and sound recordings (as in this study), that the presence of another lobster in a tank significantly increased the number of buzzing sounds produced, but that these events were also rare (mean of 3 sounds per lobster in a 30 min experimental period). Nonetheless, these authors did not perform experiments concerned with agonistic behaviours between male individuals, and in addition, the accelerometers used in Ward et al. (2011) required that the lobsters were immobilized.

In the present study, we used small accelerometers which could be attached directly on the carapace where sound production occurs (Henninger and Watson, 2005). This unobtrusive sensor permitted the lobsters to exhibit their full range of agonistic movements. In contrast to the earlier studies, we found that the number of carapace vibrations recorded with the accelerometers was very high during agonistic encounters. Indeed, we recorded a total of 422 carapace vibrations produced by 23 out of the 24 lobsters tested, with some individuals producing up to 70 carapace vibrations per experimental period (15 min total). In contrast, only 15% of these carapace vibrations were picked up by the two hydrophones (e.g. Fig. 3). This difference in detection between hydrophones and accelerometers is explained by the high attenuation of low-frequency sounds in tanks. Although low frequencies are less attenuated than high frequencies in open water, the situation is reversed in tanks when the wavelength of the sound is larger than the tank size (e.g. a 100 Hz sound has a ∼15 m wavelength). This phenomenon is well known in the acoustic community (Gray et al., 2016; Rogers et al., 2016), but sometimes misunderstood in the bioacoustic community. Nonetheless, it was recently highlighted through numerical simulations and empirical measurements. Indeed, Duncan et al. (2016) illustrated that the attenuation in a tank at 100 Hz is 10 dB higher than in open water (note that the exact number depends on the specific tank size and the source/receiver configuration). Moreover, Jézéquel et al. (2019) performed an empirical illustration of this phenomenon by comparing spiny lobster sounds in a tank and in situ. Because the high attenuation of low frequencies has been ignored in previous bioacoustic tank studies that relied on hydrophones alone, we believe that the role and importance of buzzing sounds for lobsters during agonistic encounters have been underestimated.

The detection or determination of communication amongst individual animals is a fundamental challenge in behavioural ecology (Hebets and Anderson, 2018). Communication is defined most simply as a transfer of information from one or more individuals that is observed to change the behaviour of one or more receiving individuals. Information can be transmitted and perceived in many different ways (e.g. chemically, visually, acoustically) depending on the sensory capabilities of the organisms involved. Several studies have already shown that male H. americanus use chemical signals as a means of communication to both recognize individuals and maintain dominance (Atema and Engstrom, 1971; Karavanich and Atema, 1991; Breithaupt and Atema, 2000). These same mechanisms are also known for male H. gammarus (Skog et al., 2009). Studies have demonstrated that the volume of urine released is closely linked with aggressive behaviours (Breithaupt et al., 1999) and, after the first encounter, only dominant individuals continue to release urine to maintain their dominance (Breithaupt and Atema, 2000). There is also evidence that lobsters rely on visual signals to recognize each other (Gherardi et al., 2010; Bruce et al., 2018). All these means of communication emphasize the importance of individual-level recognition of submissive and dominant individuals. For example, this would be an advantage for avoiding additional aggressive claw contact incidents that could lead to injuries and even loss of an appendage (Breithaupt and Atema, 2000).

Dominance among male lobsters also relies on their relative size differences (Scrivener, 1971). In our study, 6 out of the 12 agonistic encounters were performed with closely size-matched pairs (ΔCL<5 mm). As the encounters studied here represent examples of possible outcomes of new arrival dominance contests and not repeat encounters, the conditions should be suitable for a more important role of other signals conditioning the outcome, particularly for the encounters with closely size-matched pairs. In accordance with this, there were some intriguing patterns in the production of carapace vibrations. Indeed, we observed that submissive individuals always produced carapace vibrations, and these mostly occurred while in the claws extended pose, as well as having a broader range of duration and a higher number of carapace vibrations produced (Figs 4 and 5). In contrast, dominant individuals did not always produce a carapace vibration (e.g. Fig. 5E). We also noted that no carapace vibrations were produced by lobsters while isolated before agonistic encounters. However, examination of the carapace vibration sequences using the paradigm of AC (Gauvrit et al., 2016; Zenil et al., 2015preprint, 2018) indicated that these sequences cannot be differentiated from a random process. As stated earlier, this could be due to their non-detection by lobsters because of the high attenuation of these low-frequency sounds in the tanks (Gray et al., 2016; Rogers et al., 2016).

When looking at sequences of carapace vibrations between the two groups of encounters with different relative CLs, the more closely size-matched pairs (ΔCL<5 mm) appeared to make a greater investment in countering the strategies of their opponents. Indeed, these encounters had more carapace vibrations, which implied more effort expended to counter the opponent's reactions (Fig. 5). These preliminary results are consistent with the hypothesis that carapace vibration sequences in pairs of nearly sized-match individuals contribute to the communication of dominance, but that when size differences are larger, other signals (i.e. visual) are sufficient to establish dominance (Scrivener, 1971; Atema and Voigt, 1995; Skog et al., 2009). Interestingly, such multimodal communication is well known in terrestrial arthropods (e.g. Elias et al., 2006). However, we caution that as stated above, the vibration sequences cannot be distinguished from a random process and that there is a potential bias due to sound attenuation in tanks, as well as a small number of observations.

Our results emphasize not only the numerous technical challenges in these experiments but also the absence of knowledge about how lobsters may perceive sounds. For instance, in our study, the lack of a call-and-response pattern with carapace vibrations between lobsters was surprising. Indeed, individuals only produced vibrations when in the presence of a potential opponent, strongly suggesting their emission is context dependent. If combinatoriality (that is, the property of constructing meaning from apparently meaningless elements) is present, then the acoustic production would be considered communication if it can be shown to provoke a predictable response (Engesser and Townsend, 2019). This highlights the need to better understand how animals perceive sounds to be able to design appropriate experiments.

Lobsters cannot directly detect pressure from buzzing sounds, but they may still be able to detect the corresponding particle motion (Breithaupt and Tautz, 1990; Breithaupt, 2001; Popper et al., 2001). Indeed, a large diversity of sensory receptors dedicated to this function is known in both H. americanus and H. gammarus, including statocysts and sensory hairs (Cohen, 1955; Laverack, 1962). By considering this, Breithaupt (2001) suggested that lobsters may only be able to detect these sounds in the near-field, i.e. at distances less than a few tens of centimetres from the source. This hypothesis is consistent with the close-range communication well described in terrestrial arthropods (Raboin and Elias, 2019). Here, we did not measure or model the acoustic particle motion field in the behavioural area as this was out of the scope of the study. As a result, if the lobsters were unable to detect the buzzing sounds using particle motion, we do not know whether this is due to the specificities of tank acoustics and/or because of biological reasons. Validating (or rejecting) this hypothesis would require further work, including model and/or measurement of near-field particle motion of lobster buzzing sounds (active and reactive intensity; e.g. Zeddies et al., 2012; Jones et al., 2019), and a better understanding of the lobster sound perception system (Breithaupt, 2001).

While some studies have confirmed experimentally the role of sound production in marine crustaceans to deter predators (Bouwma and Herrnkind, 2009; Ward et al., 2011), few studies have demonstrated these sounds are used for intraspecific communication. Interestingly, stomatopods produce low-frequency sounds termed ‘rumbles’ that are similar to the lobster buzzing sounds (Patek and Caldwell, 2006; Jézéquel et al., 2018). Mantis shrimps are territorial species living in burrows, like lobsters, and their sounds might help to send signals of their presence to conspecifics to maintain territory (Staaterman et al., 2011). Spiny lobsters have also been shown to emit antennal rasps during agonistic encounters in tanks (Mulligan and Fischer, 1977), suggesting that these sounds may be used as a threat display. Snapping shrimps may also use their powerful ‘snaps’ to deter other conspecifics from their territory (Schmitz and Herberholz, 1998). During agonistic encounters, male hermit crabs produce rapping sounds by rubbing their claws against their carapace, which may be a signal of stamina (Briffa et al., 2003). In marked contrast to other marine crustacean species where behavioural responses to sounds are not yet clear, semi-terrestrial crabs (Ocypodidae) have been shown not only to produce sounds (e.g. Taylor et al., 2019) but also to respond to these sounds during intraspecific interactions (Crane, 1966; Horch and Salmon, 1969; Horch, 1975).

In our earlier study (Jézéquel et al., 2018), we found that H. gammarus produced buzzing sounds when stressed by handling. In the present study, agonistic encounters led to stressful events for both dominant and submissive individuals that resulted in the production of buzzing sounds. Thus, these sounds may be used in a similar context to the spiny lobster antennal rasps and the mantis shrimp rumbles to repel other organisms, whether conspecifics or heterospecifics (Mulligan and Fischer, 1977; Bouwma and Herrnkind, 2009; Staaterman et al., 2011). Taken together, these preliminary results suggest that male H. gammarus could use buzzing sounds, in addition to visual and chemical signals (Skog et al., 2009), as a means of intraspecific communication during agonistic encounters. However, we emphasize that our study should be repeated and include additional tests to evaluate whether these buzzing sounds really constitute communication. Other experiments should test behavioural reactions to emitted sounds as well as build an audiogram for the species associated with the quantification of particle motion (Goodall et al., 1990; Popper and Hawkins, 2018). As shown in this study, because small tanks highly attenuate buzzing sounds, these experiments should be done under controlled conditions or directly in the field (Gray et al., 2016; Rogers et al., 2016). This would also be expected to change the behavioural observations. Indeed, it is not yet known at what frequency and intensity lobsters fight for dominance under in situ conditions where escape is possible (Karnofsky et al., 1989). Finally, additional studies should address the acoustic behaviour of female lobsters during agonistic encounters, as they have also been shown to be aggressive towards conspecifics (Skog, 2009).

In this study, we have highlighted for the first time that male H. gammarus produce buzzing sounds during agonistic encounters. Notably, we showed that they only emitted sounds when placed in contact with each other, and that most of these sounds were produced after the first encounter (i.e. hierarchical status had been determined). However, we did not find clear evidence that these sounds could be used for communication between individuals. This may be due to the high attenuation of the buzzing sounds in tanks, which could prevent their perception by receivers. Other studies have suggested that these buzzing sounds could be a means of intraspecific communication in lobsters (Breithaupt and Tautz, 1990; Breithaupt, 2001). Further studies are now needed to validate this new hypothesis.

We warmly thank Céline Liret, Dominique Barthélémy and the aquariology staff of the public aquarium Océanopolis in Brest (France) for their technical support during this work. We also thank Sébastien Hervé of Université de Bretagne Occidentale for producing the illustration in Fig. 1. We thank two anonymous referees and Dr Patek for their valuable comments that improved the clarity of our manuscript.