Identification of genetically and oceanographically distinct blooms of jellyfish

Abstract

Reports of nuisance jellyfish blooms have increased worldwide during the last half-century, but the possible causes remain unclear. A persistent difficulty lies in identifying whether blooms occur owing to local or regional processes. This issue can be resolved, in part, by establishing the geographical scales of connectivity among locations, which may be addressed using genetic analyses and oceanographic modelling. We used landscape genetics and Lagrangian modelling of oceanographic dispersal to explore patterns of connectivity in the scyphozoan jellyfish Rhizostoma octopus, which occurs en masse at locations in the Irish Sea and northeastern Atlantic. We found significant genetic structure distinguishing three populations, with both consistencies and inconsistencies with prevailing physical oceanographic patterns. Our analyses identify locations where blooms occur in apparently geographically isolated populations, locations where blooms may be the source or result of migrants, and a location where blooms do not occur consistently and jellyfish are mostly immigrant. Our interdisciplinary approach thus provides a means to ascertain the geographical origins of jellyfish in outbreaks, which may have wide utility as increased international efforts investigate jellyfish blooms.

1. Introduction

The biodiversity of the seas is changing [1]. While many species have been depleted by over-exploitation [2], scyphozoan jellyfishes have become notable because ‘blooms’ of high biomass may be increasing in frequency [3] and could interfere with human use of marine resources [4]. For species with a metagenic life cycle, ‘true’ blooms [5] occur with fairly regular (annual) periodicity as a normal demographic consequence of this life cycle. The metagenic life cycle consists of populations of asexual benthic polyps that strobilate to produce numerous free-swimming ephyrae, which rapidly grow and develop into large sexually reproducing medusae. That blooms necessarily involve demographic processes and can recur persistently over hundreds of years [6], suggest population-specific dynamics. However, blooms of large medusae may also appear and disappear at some locations more rapidly than demographic processes can act (‘apparent blooms’), suggesting an important role for net oceanographic transport [7]. On a case-by-case basis, we have little idea whether the medusae that form a jellyfish bloom have common origins in local demographic processes, have originated from elsewhere, or are aggregated from a combination of sources [5]. We also do not know the geographical scales on which localities might be connected by dispersal. This dearth of information inhibits the identification of specific factors that lead to jellyfish blooms.

Spatially explicit dispersal hypotheses can be tested by comparing genetic connectivity with geographical features [8] and ocean currents [9–12]. Genetic partitioning is a consequence of many generations of separation, but can be rapidly disrupted by gene flow [13]. Genetic data, therefore, inform about long-standing contemporary barriers to dispersal [9]. At one end of the spectrum, if dispersal is low, then individuals forming a bloom should be genetically similar to each other because of shared origin and history, and genetically different from blooms originating in other localities. In this situation, blooms are a consequence of local events [14]. At the other end of the spectrum, if dispersal is high, then individuals forming a bloom may be genetically variable, with high gene flow among locations and blooms a consequence of events elsewhere [15,16].

Identifying the geographical scale of population dynamics in scyphozoans is a challenge that parallels the long-standing effort to explain population genetic structure and dynamics of other marine taxa in terms of life history [17]. The few intraspecific phylogeographic studies of jellyfishes have typically found evidence of geographically partitioned genetic structure, across large geographical distances [15,18,19]. This initially was surprising because jellyfish species with long-lived pelagic medusa were expected to be more widespread and genetically homogeneous [16] than marine organisms with short-lived pelagic phases [20–22]. However, gene flow is also influenced by biotic and abiotic factors [23] such as local and regional hydrography [21,24], behaviour [25] and environmental heterogeneity [26]. In retrospect, the asexual sessile polyps that require shallow-water benthic habitat, and the coastal environment of the free-swimming medusae could both restrict dispersal. Indeed, metagenic jellyfish medusae are more frequent in bays and close to coastlines [27,28]. Therefore, it may be important to consider the coastal landscape in addition to oceanographic currents in a multi-disciplinary approach to provide complementary perspectives on the connectivity of jellyfish populations.

For some marine organisms, estimates of gene flow from molecular analyses often correspond well with estimates of physical connectivity based on oceanographic models [8–10,18,29–31]. In this study, we include both genetics and oceanography to characterize the population structure of Rhizostoma octopus (Linnaeus 1788), a prominent jellyfish species of the northern and western coasts of Europe [32]. Rhizostoma octopus is in some ways a typical scyphozoan—it has a bipartite life history with large natatory medusa [33]—and in others atypical—it is the most northerly distributed member of Rhizostomeae, a primarily tropical order. Rhizostoma octopus is the northernmost of three species of Rhizostoma inhabiting European waters, being distributed from the southern North Sea and coasts of western UK and Ireland southwards into France. We focus our study of R. octopus on a spatial scale of a few hundred kilometres (figure 1a), with locations that are potentially connected or isolated by oceanographic currents (figure 1b).

Figure 1.

Maps of (a) sampled locations (solid circles: see table 1 for map coordinates) showing direct distances (dashed lines) and possible landscape routes (solid black lines, numbered) between locations (see electronic supplementary material, table S1), and (b) the residual flow field (21 July–20 August) for the area of study showing the connectivity of locations by oceanographic currents. The arrows indicate the magnitude (refer to the residual vector scale) and directions of residual flow.

2. Methods

2.1. Oceanographic modelling and physical connectivity

To estimate physical oceanographic connectivity between jellyfish populations in the Irish Sea and NE Atlantic where R. octopus blooms have been well described [32], we developed a Lagrangian jellyfish dispersal model. Field and experimental observations of Rhizostoma species have indicated that polyps strobilate (release ephyra) during springtime, and ephyra develop into medusae that mature by the end of the summer of the same year [34]. Our model aims to estimate how far the free-swimming forms (ephyrae and medusae) can disperse in this period of a few months.

We simulated the oceanographic currents using the three-dimensional baroclinic/barotropic Proudman Oceanographic Laboratory Coastal Ocean Modelling System (POLCOMS) [35] at a grid resolution of 1/12° in both latitude and longitude, with 20 terrain-following (sigma) layers in the vertical plane. The model was validated using altimetry data, tide gauge data, current meter data, in situ temperature time series, drogue tracks and other models that provide overlap with our model domain [36,37] (see further information in electronic supplementary material). The depth-averaged currents from the POLCOMS model were exported for subsequent input into a (offline) Lagrangian particle-tracking model (PTM) that accounts for both advective (deterministic) and diffusive (stochastic) processes to simulate jellyfish dispersal, but excludes behaviour and mortality. After a model spin-up of one month (from 1 to 31 March), 10 000 particles were released every 5 days from each sample area, and every particle was tracked until the end of the simulation (180 days from the first release date), i.e. the model was run from 1 April for 180 days. So, particles released earlier in this time-period ran for longer than those released later. For consistency across the tracked particles, we constrained the subsequent analysis of ocean connectivity to all those particles tracked for 120 days (see below). This is an appropriate timescale according to what is known about the Rhizostoma life cycle [34]. Sample areas were those where R. octopus blooms have been reported and sampled (see §2.2.). The bloom sites differed in their spatial extent, and so to provide a realistic view of the drift scenarios from each site, we seeded particles across the extent of each site. Each site had the same number of particle releases as we were quantifying the relative drift probabilities between sites with our methodology. So the particle release areas varied in size from 4 × 4 to 16 × 15 km, reflecting the size of R. octopus bloom areas identified by aerial surveys in the different locations [32]. Although, in common with most model studies [38], the Lagrangian dispersal model has not been directly validated, it provides a scientifically rigorous theoretical tool for predicting dispersal, because it is driven by hydrodynamic flowfields that have been validated.

At the end of each 180-day simulation, there were 360 000 particles in the model domain, representing 10 000 new particles introduced every 5 days. We used these data to plot particle densities (figure 2). To evaluate the relative oceanographic connectivity between locations, we estimated for a subset of the simulation period the sum of particles reaching a location after having been released from another location. For each of the first 12 releases of 10 000 particles (i.e. 120 000 particles from each of five sites; 600 000 particles in total), we had 120 days of movement data. For each of these particles at each 5-day time-step, we calculated the distance to the centre of the other four jellyfish bloom locations to which a particle might travel and counted particles travelling within 40 km of the centre (the capture distance). We used this distance to represent the movements of jellyfish from one site to another. The R. octopus bloom locations encompass the area where jellyfish are seen in aerial surveys, albeit at lower densities than in the centre. However, there are potentially many millions of R. octopus individuals in bloom sites [32], and different parts of the bloom may contribute disproportionately to reproductive success and settlement of the planulae into the benthos. Hence, it is important to note that our particle-tracking simulations do not reflect the actual numbers of jellyfish in the system, but provide estimates of relative dispersal connectivity strength. Furthermore, PTM should only be considered a ‘best estimate’ of jellyfish dispersal potential [39]. The capture distance (40 km to the centre of each destination area) provides an indication of relative rates of transport between sites, rather than absolute numbers of jellyfish being transported.

Figure 2.

Plots of particle-tracking simulations using 1/12 degree shelf grid for the five study sites. Release locations for each plot are indicated by arrows. Plots show particle densities at the end of the simulation (180 days). Model cells are colour-coded, ranging from dark blue (1–50 particles) through to dark red (more than 500 particles). For names of study sites and other place names, see figure 1; for map coordinates, see table 1. (Online version in colour.)

2.2. Collection and preservation of specimens

We sampled R. octopus between 2003 and 2005 at four locations in the Irish Sea (table 1; figure 1), including three sites where blooms consistently occur (Carmarthen Bay, CAR; Rosslare, ROS and Tremadoc Bay, TRM) and one site where blooms were documented as being less consistent (Gormanstown, GOR) [28,32]. In 2008, we collected samples from La Rochelle in France (LAR), which also has consistent blooms of R. octopus [32]. Rhopalia and/or gonads were preserved in a greater than 95 per cent ethanol and frozen at −20°C. Samples were from scyphomedusae captured at sea except for CAR and GOR where samples were from stranded jellyfish.

Table 1.

Sample locations (country, site), site acronym, map coordinates (N,W), sample sizes (by year), h (gene (haplotype or allelic) diversity) and π (nucleotide diversity) for mitochondrial cytochrome c oxidase subunit I (COI) and nuclear calmodulin (CaM).

				COI			CaM
location	site acronym	latitude (N)	longitude (W)	n (nyear, nyear)	h ± s.d.	π ± s.d.	n (nyear, nyear)	h ± s.d.	π ± s.d.
Ireland
Gormanstown	GOR	53°40′46.70″	6°11′27.71″	72005	0.667 ± 0.160	0.00337 ± 0.00249	72005	0.824 ± 0.098	0.00293 ± 0.00218
Rosslare	ROS	52°17′34.06″	6°20′39.55″	14 (32003,112004)	0.791 ± 0.089	0.00629 ± 0.00382	14 (32003,112004)	0.944 ± 0.021	0.00470 ± 0.00287
Wales
Carmarthen Bay	CAR	51°40′32.65″	4°30′3.28″	14 (122003, 22004)	0.879 ± 0.058	0.00666 ± 0.00401	14 (122003, 22004)	0.918 ± 0.027	0.00517 ± 0.00311
Tremadoc Bay	TRM	52°52′33.64″	4°11′0.54″	14 (72003, 72004)	0.396 ± 0.159	0.00144 ± 0.00123	13 (62003, 72004)	0.843 ± 0.045	0.00280 ± 0.00191
France
La Rochelle	LAR	46°18′0.21"	1°22′18.06"	142008	0.879 ± 0.052	0.00326 ± 0.00223	142008	0.828 ± 0.062	0.00522 ± 0.00313

2.3. DNA extraction, amplification and sequencing

We purified genomic DNA following a CTAB protocol [40]. We amplified a fragment of the mitochondrial gene cytochrome c oxidase subunit I (COI) by PCR using primer LCOjf (5′-ggtcaacaaatcataaagatattggaac-3′) with either HCO2198 [41] or Acro_HCO_611 (5′-agcagggtcgaagaaagatgtatt; from K. Bayha). We PCR-amplified a portion of the nuclear calmodulin gene (CaM) using new primers Ms_LCaM_100 (5′-cgacaaagatggmgayggyac) and Ms_HCaM_426 (5′-gcttchckratcatytcrtc) designed using preliminary sequence data (by K. Bayha, 2010, unpublished data) obtained using published primers [42]. PCRs contained 1 μl of template DNA, 5 µl 10X PCR buffer, 0.2 mM each dNTP (GeneAmp dNTP mix with dTTP, Applied Biosystems Inc., Bethesda, MD, USA), 2.5 mM MgCl₂, 0.5 units (U) of Amplitaq (Applied Biosystems) and 0.5 μM each primer (Invitrogen, Carlsbad, CA, USA) in a 50 μl volume. Thermal cycling began with up to six holds of 94°C for 3–8 min, 49–57°C for 2 min, 72°C for 2–3 min, 94°C for 4 min, 50–56°C for 1–2 min, 72°C for 2–2.5 min and then 33–38 cycles of 94°C for 30–45 s, 50–55°C for 45–60 s, 72°C for 60–90 s depending on the primer pair. All PCRs completed with a 10 min extension at 72°C and were terminated by cooling to 4°C. (For specific PCR protocols, see electronic supplementary material.)

PCR products were directly sequenced, or sequenced after cloning with Invitrogen's (Carlsbad, CA, USA) TA cloning kit with TOP 10 competent cells at the high-throughput genomics unit (University of Washington). We edited and compiled forward and reverse sequences, keeping multiple CaM haplotypes from a single animal as separate contigs, in Sequencher v. 4.9 (GeneCodes Corp., Ann Arbor, MI, USA), which we then aligned in ClustalX v. 2.0 [43] using default settings. We identified CaM alleles by examining homozygotes, heterozygotes polymorphic at only one position, and by subtracting clones from direct sequenced trace-files with more than one polymorphism in Se-Al v. 2.0a11 [44]. We then used these known alleles to infer haplotypes from the remaining genotypic data by using Bayesian coalescent reconstruction allowing for recombination in Phase v. 2.1 for Mac [45–47]. Sequences were deposited in GenBank under accession numbers HQ425339–HQ425479. DNA substitution models for each locus were estimated using Modeltest v. 3.7 [48] and/or transition : transversion ratios from Arlequin v. 3.5.1.2 [49]. We constructed statistical parsimony networks for the haplotypes or alleles using TCS [50], according to recommendations [51].

2.4. Genetic analyses

We tested for population expansion with Tajima's D, Fu's Fs and the fit of the mismatch distribution to a demographic expansion using Arlequin v. 3.5.1.2 [49]. To estimate coalescence time t from the parameter τ (mutation-scaled time to the expansion) [34], we used a rate of substitution estimated for hydrozoans (Cnidaria) of 6.54 × 10⁻⁹ substitutions per site per year (0.65% divergence per million years) [52] and a rule-of-thumb rate of 0.5–2% divergence per million years for COI [53,54], and assumed a 1-year generation time, which may be an underestimate for populations with overlapping generations. Because μ and generation time are approximations, it is important to note that the estimated coalescence time t provides only a rough guide.

We calculated Φ_ST values (10 100 permutations), linearized F_ST and Slatkin's M, and conducted exact tests of population differentiation (Markov chain length 10 000 steps) and analysis of molecular variance (AMOVA) using arlequin [49]. Significance levels were corrected for multiple tests using the sequential Bonferroni procedure [55].

IBD tests (http://ibdws.sdsu.edu/~ibdws/ [56]) considered effects of three distances. First, Euclidean or direct distances (EDs), as a simple approximation of the shortest route between locations. Second, ‘oceanographic distances’ (ODs) calculated as 1/(1 + oceanographic connectivity) (oceanographic connectivity values are the number of particles that had dispersed from one location to the other by the end of the particle-tracking simulation; see §2.1.). Third, ‘landscape distances’ (LDs) measured along coastlines plus open water, representing dispersal routes through coastal zones thought to be the primary habitat for R. octopus and including shallow-water benthic habitat essential for inter-generational dispersal including polyp settlement and strobilation. Ship-based and aerial surveys show that R. octopus medusae occur mainly in bays, and when outside of bays medusae usually are close to the shore with only the occasional individual observed in open water [27,28]. LDs between locations on the same coastline were estimated as distances along-shore. LDs between locations also separated by open water were estimated by combining along-shore distances with possible short across-sea routes. The short across-sea routes are consistent with observations of R. octopus or other species of jellyfish in the Irish Sea [28], or bracket similar possibilities across the English Channel, but, because ephyrae may be too small and medusae may occur too deep to be observed from ships or aeroplanes, do not exclude other possible across-sea routes. Distances were measured using Google Earth v. 6.0.2 (http://www.google.co.uk/intl/en_uk/earth/index.html). For LDs, a heuristic approach was used to test various possible routes in IBD (electronic supplementary material, table S1).

Genetic barriers for COI, CaM and COI + CaM combined were estimated with barrier v.2.2 [57].

2.5. Testing among models of connectivity

Pairwise F_ST is commonly used to estimate the degree of migration, but migrate maybe more appropriate, as it uses Bayesian inference and coalescence theory to directly estimate migration rate and to calculate probabilities of explicit models of connectivity among populations [58]. migrate has advantages in that a range of migration models can be evaluated against each other, and is robust to incomplete sampling [58,59]. We used Migrate-n v.3.2.6 [58] to test among models of connectivity derived from observed jellyfish occurrences, oceanographic modelling and simple population genetic hypotheses:

• (1) ‘Full model’: full symmetric migration model.

• (2) ‘Panmixia model’: all samples are of the same population.

• (3) ‘GOR sink model’: GOR sample-set receives but does not contribute migrants.

• (4) ‘Particle-track model’: model based solely on oceanographic connectivity.

• (5) ‘Four population full model’: samples from CAR and ROS are assumed to be of the same population, as suggested by oceanographic connectivity.

• (6) ‘Combination model’: combination of (3) and (5).

The ‘full’ and ‘panmixia’ models provided simple models for comparison. The ‘GOR sink model’ was based on the observation that blooms do not occur here consistently [28,32], and tested the possibility that the area may be a sink. The ‘particle-track model’ tested the possibility that jellyfish connections were determined solely by oceanographic forcing (figure 2)—migration was allowed only between locations connected by short-term oceanographic connectivity. The ‘combination model’ (model 6) allowed migration to occur between locations unconnected by short-term oceanographic connectivity, but considered the location without recurring blooms as a sink (model 3) and locations with consistent blooms connected by oceanography as one population (model 5). Model 6, therefore, combined what we knew from published field observation with the pattern of connectivity suggested by oceanographic modelling, yet allowed for migration.

We estimated the parameters θ (mutation-scaled population size), M (mutation-scaled rate of migration: m/μ) and x (multiplier that depends on the inheritance of the data and on ploidy) × N_e (effective population size) × μ (rate of mutation) by Bayesian inference [60]. We calculated the number of migrants per generation (xN_em) by multiplying together the modal values of θ and M from the posterior distribution of these parameters as inferred by Bayesian analysis (with the 2.5 and 97.5 percentiles of the posterior distributions reported as 95% CIs). Settings selected following preliminary trials were: θ prior set to a maximum of 0.1, M prior set to a maximum of 10 000, Metropolis sampling, 100 K recorded steps (with an increment of 100 between recorded steps) and burn-in of 500 K steps. We ran five replicate analyses per model of connectivity with static heating (four chains using default settings) and exponential priors. To decide on the ‘best’ model, the ratios of marginal likelihoods for each of the connection models were compared [58]. The best model was rerun for 15 replicates and for five times more recorded steps to check for consistency of parameter estimates.

3. Results

3.1. Oceanographic modelling and physical connectivity

We found that few particles drifted between TRM and CAR, and that connections between locations were asymmetrical (figure 2). For example, particles drifted away from GOR, particularly to the south, with particles crossing to Wales. Also, particles drifted from CAR to ROS, following a strongly directional residual flow field (figure 1b). For releases from TRM and LAR, particles did not reach any of the other locations because residual currents were relatively weak in these locations. Simulated jellyfish released from ROS and GOR dispersed along the southern coast of Ireland and into the Celtic Sea (figure 2) where there are strong residual currents (figure 1b). A large proportion of the simulated jellyfish released from CAR became entrained in the baroclinic system associated with the Celtic Sea front, crossing to the southern coast of Ireland, with the remaining simulated jellyfish population dispersing within CAR and across the Bristol Channel (figure 2).

3.2. DNA sequence variation

Of 566 nucleotides of COI that we sequenced from 63 medusae, we found 23 variable positions, and 20 haplotypes (figure 3a). We compared the COI sequence data with those of R. pulmo in GenBank to confirm that these formed separate monophyletic groups (see electronic supplementary material, figure S1). The mean (±s.d.) number of pairwise differences among R. octopus sequences was 3.51 ± 1.81 nucleotides (0.6% ± 0.3%); all differences were transitions. The best-fit models (HKY and HKY + I) apply equal substitution rates for all sites so we performed subsequent calculations in arlequin with COI using pairwise sequence difference. All sets of samples contained multiple genetically distinct medusae (table 1). TRM samples had the lowest value of h (haplotype diversity) at 0.396 ± 0.159 (s.d.), whereas h for samples of all other locations ranged between 0.667 and 0.89 (table 1).

Figure 3.

Haplotype networks with the area of each shape proportional to the frequency with which that haplotype was observed, small empty circles represent unsampled haplotypes. Each branch represents a single nucleotide substitution. (a) COI haplotype network. Emanating from a central haplotype are three clusters of haplotypes found in the Irish Sea: Cluster I (2 haplotypes: I_a and I_b), Cluster II (5 haplotypes: II_a-e), Cluster III (7 haplotypes: III_a-f) (b) CaM allele network. The area of each shape is proportional to the frequency with which that allele was observed (smallest circle, n = 1; largest circle, n = 32), small empty circles represent unsampled haplotypes. Each branch between any two shapes represents a single nucleotide substitution.

From 62 medusae, we sequenced 567–621 nucleotides (aligned length 621 positions), which included both exon and intron fragments of the CaM gene. Thirty-three alleles were reconstructed (figure 3b), of which 16 had uncertain phase (less than 0.95 probability) at between one and three positions. The 20 positions with unambiguous phase showed 13 transitions, six transversions and one indel. Applying the Kimura two-parameter model of nucleotide substitution, the mean (± s.e.) number of pairwise differences between alleles was 2.59 ± 1.40 nucleotides. The h ranged from 0.824 to 0.944, and π from 0.00280 to 0.00522 and was lowest for TRM samples (table 1).

Network analysis (figure 3a) of COI haplotypes showed that none of the LAR haplotypes were found in any Irish Sea location. However, all haplotypes in the Irish Sea can be grouped into three clusters that radiate out from one of the LAR haplotypes. Cluster I consisted of two haplotypes and were predominantly found in TRM (80%), with one GOR and two ROS samples. Cluster II consisted of five haplotypes and was mostly of CAR samples (54%), and included ROS (31%) and just two TRM samples. Finally, Cluster III was the largest group with seven haplotypes, including samples from ROS (38%), CAR (33%) and GOR (29%). By contrast, half of the CaM alleles were found in multiple locations, with four of the high frequency alleles occurring in all locations (figure 3b).

3.3. Genetic analyses

Tajima's D was not significantly different from zero for COI and CaM (−0.879, p = 0.202 and −0.751, p = 0.262, respectively), while Fs results were mixed (−6.755, p = 0.011 and −5.446, p = 0.045, for COI and CaM, respectively). Mismatch distributions for COI and CaM did not differ significantly from a model of sudden expansion (θ₀ = 0.000, θ₁ = 11.534, τ = 4.623, p = 0.166 and θ₀ = 0.016, θ₁ = 10.557, τ = 3.461, p = 0.964, respectively; electronic supplementary material, figure S2). The two COI calibrations date the time of expansion to approximately 292–624 ka (95% CI boundaries: 104–472 kyr and 223–1011 kyr).

Most samples drawn from different locations were strongly differentiated from each other (see the electronic supplementary material, table S2): ^COIΦ_ST calculated for all possible pairs of samples ranged from 0.03 to 0.75, with eight of 10 values being greater than or equal to 0.28; differentiation for three locations also was evident for the nuclear marker, with ^CaMΦ_ST ranging from −0.02 to 0.13. There was no statistical difference between specimens collected at different times in TRM (2003 versus 2004: ^COIΦ_ST = 0.111, p = 0.459 ± 0.005)—temporal variation was not tested at other sites owing to small sample sizes (table 1). For the COI, but not for the CaM data, the TRM sample-set was the most distinct, with significant genetic differentiation from all other Irish Sea samples. AMOVA further supported these results (see the electronic supplementary material, table S3).

Both simulated jellyfish drift and gene flow (figure 4a–c) qualitatively identified the same two major connections (GOR and ROS, and CAR and ROS), although quantitative estimates of connectivity differed between methods. Analysis of combined COI and CaM data identified two barriers to gene flow: one between LAR and the Irish Sea sites, and the other between TRM and the other Irish Sea sites (figure 4d). Separate analyses for COI and CaM data confirmed the same barriers.

Figure 4.

(a) Diagrams summarizing oceanographic and (b,c) genetic connections, (d) genetic barriers and (e,f) best migrate connection models. Connectivity according to (a) particle drift as determined by circulation models, and (b) COI and (c) CaM genetic distances estimated as Slatkin's M (note that M = Nm for haploid COI and 2 Nm for diploid CaM data). (d) Genetic barriers (triple lines) between locations were identified using Monmonier's maximum difference algorithm from COI and CaM DNA sequence data. migrate connection models for (e) COI and (f) CaM data show the number of migrants per generation (with 95% CI in italics) for the ‘Combination Model’ (table 2) for which GOR receives but does not export migrants, and ROS and CAR are a single population. Location labels: G, Gormanstown; T, Tremadoc Bay, C, Carmarthen Bay; R, Rosslare and L, La Rochelle. Line weights illustrate the approximate relative degree of connection; arrows indicate the direction of connection. For clarity, connections with values less than 1 are not shown.

For COI data, genetic distances among sites correlated better with various LDs (e.g. r² = 0.189–0.52, p = 0.162–0.021 for log-transformed Slatkin's M and log-transformed geographical data) than with EDs (r² = 0.124, p = 0.686; see the electronic supplementary material, table S4 for further examples). For CaM linearized F_ST, a stronger correlation was found for log-transformed LDs (r² = 0.043, p = 0.243) than for log-transformed EDs (r² = 0.153 with p = 0.281; see the electronic supplementary material, table S4 for results of other comparisons). However, IBD tests comparing ODs revealed a strong correlation between CaM log-transformed linearized F_ST data and log-transformed ODs: r² = 0.734 with p = 0.0150. By contrast, there was no relationship between ODs and COI data (r² = 0.336 with p = 0.132).

3.4. Testing among models of connectivity

Parameter estimates of migrate analyses reached convergence (see the electronic supplementary material, table S5). With COI data, θ indicated larger N_e for LAR and ROS + CAR, compared with GOR and TRM, but this was less evident with CaM (electronic supplementary material, table S5). Independent model testing for COI and CaM found that the ‘combination model’ had the highest probability for both datasets (table 2). By contrast, the ‘particle-track model’ based on short-term oceanographic modelling fitted the genetic data most poorly.

Table 2.

Comparison of connectivity models using log Bayes factors (LBFs) and Bezier approximated log marginal likelihoods (Bezier lmL). The ‘combination model’ is a combination of the ‘GOR sink model’ and the ‘four-populations model’.

model	Bezier lmL	LBF	rank	model probability
COI
full	−1038.129	−2.542	4	0.050
panmixia	−1041.891	−6.304	5	0.001
GOR sink	−1036.539	−0.952	2	0.247
particle-track	−1059.946	−24.360	6	0.000
four populations	−1037.923	−2.336	3	0.062
combination	−1035.587	0	best	0.640
CaM
full	−1087.348	−5.740	5	0.002
panmixia	−1083.265	−1.657	3	0.131
GOR sink	−1084.724	−3.116	4	0.031
particle-track	−1215.256	−133.648	6	0.000
four populations	−1083.146	−1.5383	2	0.147
combination	−1081.608	0	best	0.688

4. Discussion

The processes driving spatio-temporal dynamics of gelatinous zooplankton blooms are enigmatic [4,61]. Studies on a wide range of other organisms have illustrated the effectiveness of combining genetic and oceanographic techniques, but here for jellyfish, while we found evidence that some populations are retained locally, the links between modern ocean currents and patterns of gene flow were mixed.

Rhizostoma octopus is genetically differentiated at the scale of a few hundred kilometres, with genetically distinct blooms at different locations (see the electronic supplementary material, table S2). While there were more nuclear alleles than mtDNA haplotypes (figure 3), the signal of genetic differences between locations was less pronounced for CaM than for COI, as expected for a ‘lagging’ indicator with a larger effective population size [62] and slower mutation rate [63]. Genetic structure did not fit simple isolation-by-ED, but was better explained by longer and more circuitous LD routes (see the electronic supplementary material, table S3). This is consistent with the distribution of shallow-water benthic habitat necessary for polyps and occurrence of R. octopus medusae mainly in bays or close to shore [27,28].

Our oceanographic simulations of jellyfish dispersal showed that while currents connected some locations (GOR and ROS, and CAR and ROS), at least two were relatively isolated (no particles from either LAR or TRM dispersed to any of the other locations, and relatively few reached TRM from elsewhere; figures 2 and 4a). LAR and TRM were also locations isolated by genetic barriers (figure 4d). One explanation could be that LAR samples were sampled in a different year (2008 instead of 2003–2005). However, the TRM samples were collected in the same period of time as the other Irish Sea samples; so isolation by currents is a likely explanation for the genetic distinction of this population. Because recurring blooms are documented for this location, we suggest that these may be caused by established populations of polyps located close to where the blooms are observed to occur.

However, blooms may occur only occasionally at one location in this study where medusae have been regularly sighted [32]. Our genetic analyses suggest that one such site, GOR, is more likely a sink than a source of medusae for the other locations (table 2). This may explain some of the mixed results; for example, the relatively low COI genetic connectivity between GOR and ROS (figure 4b), despite the high degree of particle dispersal predicted between these locations (figure 4a). If there is no local benthic polyps population near GOR, then medusae in GOR must have originated elsewhere, possibly from populations not sampled in this study. For example, the prevailing currents would deliver immigrants from the north (figure 1b). The medusae sampled at GOR had been stranded there, so perhaps few migrants would survive long enough to reach the other Irish Sea locations further south. Clarification of this hypothesis requires additional sampling of blooms occurring further north (e.g. R. octopus is regularly sighted and occasionally blooms at the Solway Firth [32]).

Qualitatively, simulated jellyfish drift and genetic distances (figure 4a–c) seemed to identify the same major connections (GOR and ROS, and CAR and ROS). Indeed, IBD analysis found a significant correlation between the nuclear CaM genetic distances (figure 4c) and ODs (figure 4a). However, ODs were not related to the COI genetic distances (figure 4b), and did not explain all the patterns of genetic connectivity estimated by MIGRATE (figure 4e,f). Perhaps mtDNA haplotypes are better as indicators of barriers to dispersal, while it takes more time and/or greater isolation before significant differentiation of nuclear alleles becomes evident (see discussion by Puritz & Toonen [64]). Another possibility is that various genetic loci and oceanographic patterns reflect different aspects of dispersal such as historical versus modern and/or rare versus mean routes [13,65]. For example, the estimated time of expansion based on COI data at 292–624 ka is far earlier than the last glacial period that ended about 20 ka [66]. This genetic signal is therefore in part that of an older expansion, pre-dating the current Holocene and more coincident with interglacials occurring 200–630 ka [67]. Rhizostoma octopus populations probably have been periodically expanding northwards and contracting southwards in response to the recent ice-age cycles [66], so we suggest that the current partitioning of COI haplotype diversity may in part reflect palaeocurrents and dispersal routes [68], as well as current isolation.

In the marine environment, in addition to barriers to connectivity (e.g. circulation patterns), genetic differentiation may also arise through environmental heterogeneity affecting survival, or genetic drift. Cryptic species, intraspecific phylogeographic structure and local adaptation suggest restricted gene flow is commonplace in Scyphozoa [15,18,19,69–71; this study] and could arise quickly at modest spatial scales [19; this study]—within a few hundred kilometres. We suggest that other mechanisms not tested in this study also may influence jellyfish dispersal. For example, jellyfish are capable of directional swimming [72,73], and may orientate their movements to remain in embayments and avoid being advected across open water. The ability to swim against flows might be expected of adult medusae, which for this species are nekto-plankton [74]. Future studies tracking individuals with biologgers to test their ability to orientate and swim against currents are feasible [75]. Another mechanism might be natural selection against immigrant phenotypes [26]. That R. octopus is not uniformly observed across its range, but concentrated and persistently found at certain coastal sites, may be a result of specific environmental requirements for the growth and survival of polyps. Active habitat selection has been suggested for other species [76]. Differentiating among these alternative explanations would require experiments in the laboratory to determine settlement/habitat preferences, and reciprocal translocation experiments in the field to detect evidence of selection.

4.1. Summary of wider implications

The current debate about whether human-mediated environmental changes are likely to result in future changes of gelatinous zooplankton biomass [3,4,61,77] is hampered by a general lack of knowledge of the population genetic structure of jellyfish blooms. Our study extends recent efforts to redress this dearth of data for jellyfish species [15,16,18,19,78,79]. Like prior studies, we found partitioning of genetic variation in metagenic species across relatively moderate distances. The appropriate scale for considering the population dynamics of this prominent Atlantic species is as little as a few hundred kilometres. Because this spatial scale is consistent with the observation of gelatinous zooplankton blooms often being defined by fine- rather than large-scaled physical features [7], we suggest that a relatively small to moderate geographical scale may also apply for some other species.