Coordinated gas release among the physostomous fish sprat (Sprattus sprattus)

Study area

The study was carried out in Bunnefjorden, Norway. The fjord froze over from January to April and we here analyze data from ice-free conditions in early winter (12 Nov to 2 Dec 2009). Bunnefjorden is a 150 m deep inner branch of the Oslofjord, and a 57 m deep sill at the entrance restricts water exchange with the outer part of the fjord. Klevjer and Kaartvedt²⁴ provide a map of the study area. The fjord branch normally becomes hypoxic in the lower part of the water column. During the current study, oxygen contents were 2–3 ml l⁻¹ between 15 and 60 m, while waters below 70–80 m were severely hypoxic and devoid of fishes⁹.

Studies of overwintering sprat have been undertaken in Bunnefjorden during several winters, and the biology of sprat as well as the identity of the main acoustic targets in the fjord are well established^9,18,24. In the winter of the current study, catches from 33 trawl samples were dominated by sprat; with ~ 40 times higher catches than the next most abundant species, herring (Clupea harengus)⁹.

Study design

Solberg and Kaartvedt⁹ and Solberg et al.¹⁸ provide details on methods, and we here only give a summary of the acoustic setup. In short, upward-looking Simrad EK 60 echosounders kept in pressure-proof casings were deployed at the bottom (150 m) and in buoys (80 and 30 m) for enhanced resolution in shallower part of the water column. Cables for electricity and transfer of data to a PC on shore enabled continuous operation of the systems. We here use the data from the shallowest echosounder (200 kHz) that provided superior resolution in near-surface water, though did not cover the full depth range of the population distribution. Echograms from the deeper located echosounders covering the whole (inhabited) water column and showing the full diel population behavior are given in Solberg and Kaartvedt⁹ and Solberg et al.¹⁸.

Records of gas release

Released gas appeared as ascending lines in the echogram (Fig. 4). We quantified the release as explained by Solberg and Kaartvedt⁹. We only included ascending traces connected to the acoustic record of a fish, but without enumerating the release per individual fish. Since the same fish may release several bursts of bubbles within a short time interval, we here pooled any sequences of gas release within a 10-s period as one event. This procedure will also exclude cases with several different individuals releasing bubbles in the course of this short time interval, yet we chose this conservative approach not to generate an artificial high connection of gas releases between the fishes.

Analyses of data

The frequency of gas releases varied with time, both within a day and between the weeks. Such patterns compare to service systems like call centres and hospital emergency rooms²⁵ that can be modelled as a Poisson process^26,27. We therefore started our analysis with the statistical procedure suggested by Brown et al.²⁸ in their influential analysis of the call dynamics in a banking call centre. The first step is to subdivide the day into time intervals, which are short enough to consider event rates as approximately constant. Here we chose to investigate alternative periods of respectively 1, 5, and 30 min, as well as 1, 2, 4, and 6 h. At the longest interval, the peaks in the gas release intensity are expected to be the result of a non-constant Poisson parameter, and therefore more likely to induce rejection of the null-hypothesis of a homogenous random process. In contrast, we expect to find higher concordance with a random process for the short intervals of 5 min. In assessing connectivity among gas bubble releases, we formulate a new model allowing for a formal test of non-randomness (summarized in Fig. 5). We name this approach the simulated connectivity test (S-CON test), which we implemented in R²⁹, with the code being available in the Supplementary appendix.

If there is a common physiological reason or some form of communication among sprat, a burst of gas release is likely followed by subsequent releases. Thus, it is reasonable to assume that the total number of releases within a short time interval like 30 s would be effective in detecting dependencies between the releases. We therefore define the concept of connectivity as follows:

Let the gas be released at times T₁, T₂, …, T_n and define the connectivity at each event as the numbers of records within the following 30 s. The average connectivity in any considered window of the investigated time-period (for example a window of 1 h) is defined as the average connectivity of all cases of connectivity within the considered window (see also Fig. 5).

In order to test the null hypothesis of no dependency between gas bubble releases, we compare the measured average connectivity in the data set with the simulation of 1000 random placements of the total number of observations in a given time window. For example, if we consider a window of 30 min with 15 release events having an average connectivity of 2.1, we performed 1000 random placements of 15 points between 1 and 1800. In this way, we get 1000 simulated values of the average connectivity, from which we pick out the critical 95th percentile, following the common significance level of 0.05 in biology. If the observed average connectivity is larger than this critical value, we reject the null-hypothesis and conclude that the releases of gas bubbles are dependent random variables. Thus, if our example obtains a critical value of 1.7, the null-hypothesis of random arrival times of bubbles is rejected (because the observed value of 2.1 is larger than the critical value of 1.7).

Since a dependency between the fish will induce a higher concentration of release events than produced by random releases, we expect the average connectivity to be quite sensitive to the alternative hypothesis of dependent arrival times. Also, note that the concept of connectivity has a combinatory nature, so we need only require that the considered window contains at least two releases of gas bubbles. In contrast, alternative approaches using Kolmogorov–Smirnov tests²⁸ are based on the cumulative distribution function and therefore require at least five observed bubble releases.

To test the dependency of the results on the chosen time interval, we also ran the analysis using connectivity intervals of 25 and 35 s, which revealed some variability to the estimates of non-random bubble release (Fig. 3) but did not influence the general pattern. We also tested whether the interval within which we consider subsequent bubbles to be part of one single release event influences our results. The more we consider sequential bubbles to be independent of each other, i.e. their own release event, the higher the proportion of non-random gas release and vice versa.

Fish abundance

To exclude the possibility that apparent connectivity would be a mere result of fluctuating fish abundance, we tested whether the number of released bubbles is a function of fish biomass. For this, we first calculated the total number of gas release events within 30-min periods. We then compared these values to the summed surface integrated acoustic scattering coefficient (SA) for the same periods and for the same depth interval (upper 30 m), assuming that the integrated scattering coefficient (SA) serves as a proxy for the total fish biomass⁹. We filtered the scattering data to remove noise from non-biological sources prior to use. Both variables were log-transformed prior to analysis. We then fitted a linear model of the two variables using generalized least squares. To account for temporal autocorrelation in the data, we also included a correlation structure of type 1 (corAR1). The analysis was done in R²⁹ using the nlme package³⁰.

Ethics declarations

Live animals (fish) were not used in this study.

Source: Ecology - nature.com