Marine stepping‐stones: Connectivity of Mytilus edulis populations between offshore energy installations

Abstract Recent papers have suggested that epifaunal organisms use artificial structures as stepping‐stones to spread to areas that are too distant to reach in a single generation. With thousands of artificial structures present in the North Sea, we test the hypothesis that these structures are connected by water currents and act as an interconnected reef. Population genetic structure of the blue mussel, Mytilus edulis, was expected to follow a pattern predicted by a particle tracking model (PTM). Correlation between population genetic differentiation, based on microsatellite markers, and particle exchange was tested. Specimens of M. edulis were found at each location, although the PTM indicated that locations >85 km offshore were isolated from coastal subpopulations. The fixation coefficient F ST correlated with the number of arrivals in the PTM. However, the number of effective migrants per generation as inferred from coalescent simulations did not show a strong correlation with the arriving particles. Isolation by distance analysis showed no increase in isolation with increasing distance and we did not find clear structure among the populations. The marine stepping‐stone effect is obviously important for the distribution of M. edulis in the North Sea and it may influence ecologically comparable species in a similar way. In the absence of artificial shallow hard substrates, M. edulis would be unlikely to survive in offshore North Sea waters.


| INTRODUC TI ON
The past decades have witnessed a steady advance of artificial structures into the marine environment, commonly referred to as "ocean sprawl" (Bishop et al., 2017;Firth et al., 2016). These artificial hard substrates include nearshore elements such as jetties, dikes and beach groins, as well as offshore structures such as oil and gas platforms, wind turbines and buoys. Growth of hard-substrate flora and fauna on them is typically abundant (Coolen et al., 2018;De Mesel, Kerckhof, Norro, Rumes, & Degraer, 2015;Krone, Gutow, Joschko, & Schröder, 2013). An increase in connectivity for these species is one of the many kinds of effects these structures are expected to have on natural ecosystems (Adams, Miller, Aleynik, & Burrows, 2014). Artificial hard substrates can act as stepping-stones; that is, patches of natural habitat that are normally unconnected may now be connected through migration in two or more generations (Fowler et al., 2019). Marine stepping-stones can be important for the distribution of native species such as Caryophyllia smithii , Porifera sp. (van der Molen et al., 2018) and Desmophyllum pertusum (Henry et al., 2018) and most species present on offshore artificial reefs in the North Sea are indigenous for the area (Coolen et al., 2018;De Mesel et al., 2015). However, the stepping-stone effect might also contribute to a faster distribution of nonindigenous species (Adams et al., 2014;IPIECA, 2010;Macreadie, Fowler, & Booth, 2011).
A recent review of priority questions that need to be addressed in the field of decommissioning of offshore structures in relation to marine ecology emphasized the need to study the stepping-stone effect (Fowler et al., 2019). The prediction of increased connectivity due to the presence of a network of artificial hard substrates has so far only rarely been tested directly using genetic markers. A study conducted on the limpet Patella caerulea around Italy found significant population structure for microsatellite markers (Fauvelot, Bertozzi, Costantini, Airoldi, & Abbiati, 2009).
Significant population structure, and hence only low connectivity, among artificial hard substrates has also been observed in several coral species in the Gulf of Mexico (Atchison, Sammarco, & Brazeau, 2008;Sammarco, Brazeau, McKoin, & Strychar, 2017). In contrast, Sammarco, Brazeau, and Sinclair (2012) showed low levels of differentiation between coral species on artificial structures in the Gulf of Mexico. Furthermore, the polychaete Pomatoceros triqueter showed no genetic differentiation among populations from 11 offshore gas platforms in the Adriatic Sea, suggesting that connectivity might be very high in this case (Fauvelot, Costantini, Virgilio, & Abbiati, 2012). In general, sample acquisition from offshore energy installations is highly challenging, especially in the North Sea, as scientists need to build good relationships with several offshore companies. To work on offshore energy installations, scientists require permissions from the respective installation operators, diving companies and vessel companies, manifesting in various contracts. Sample collectors need to hold commercial diving and offshore safety certifications, usually being on standby for months in a row to be available for last minute planned operations and then, when finally offshore, to be allowed to enter the water to acquire samples.
An alternative approach to study connectivity among artificial hard substrates is biophysical modelling of oceanographic currents, tracking the fate of particles in calculated simulations (Hufnagl et al., 2016). For example, the dispersal range of the bryozoan Watersipora subtorquata was modelled to include some platforms in the Santa Barbara Channel while other installations were out of reach for the species (Simons et al., 2016). Connectivity between coastal populations of moon jellyfish Aurelia spp. increased due to the introduction of platforms in the Adriatic Sea (Vodopivec, Peliz, & Malej, 2017). Using artificial species representing a native and non-native species, Adams et al. (2014) showed that in two generations, larvae were able to disperse between Scotland and Northern Ireland due to the introduction of artificial structures at dispersal boundaries.
Furthermore, a congruence may be detected only at larger spatial scales but not at smaller scales (Foster et al., 2012;Johansson et al., 2015) or vice versa (Jorde et al., 2015).
Abundant communities of plant and animal taxa grow on offshore artificial hard substrates in the North Sea. Here, we chose to study blue mussels, Mytilus edulis, as they are intertidal to shallow subtidal and thus fully rely on artificial substrates offshore; in the North Sea, they do not occur on sea floor habitats such as permanently submerged rocks or gravel (Coolen et al., 2018). Mytilus edulis occurs naturally along shorelines of the northeast Atlantic from the high Arctic to the English Channel (Skibinski, Beardmore, & Cross, 1983). It hybridizes in stretched out mosaic hybrid zones with Mytilus trossulus, which inhabits the Baltic Sea in the east and with Mytilus galloprovincialis in the south (Fraïsse, Roux, Welch, & Bierne, 2014;Hilbish et al., 2012). Populations along the North Sea shores are M. edulis (Bierne et al., 2003;Luttikhuizen, Koolhaas, Bol, & Piersma, 2002) and therefore we expected the same for the offshore North Sea. Mytilus edulis has a pelagic larval stage of 16-70 days after which the larvae metamorphose to the pediveliger stage during which it is capable of settling on suitable substrates (Bayne, 1965;Filgueira, Brown, Comeau, & Grant, 2014;van der Molen et al., 2018). In the southern North Sea, highest concentrations of M. edulis larvae are observed between March and July (P. Kamermans, Wageningen Marine Research, personal communication).  suggested that species with a pelagic larval stage of up to 10 weeks are unlikely to colonize far offshore objects in the North Sea without a source subpopulation located on an upstream intertidal object. However, many platforms and other shallow objects such as wind farms and navigational buoys exist in the North Sea  Figure 1), making it likely that coastal larvae reach a suitable artificial substrate at the time the pediveliger stage begins, after which its offspring may be able to reach the far offshore object.
Here, we present results from particle tracking models (PTMs) and validate them using multilocus genetic data for M. edulis for the southern North Sea region, an area with abundant offshore human activity. We hypothesized that anthropogenic offshore installations enable M. edulis to extend its distribution to areas that would typically not be accessible for coastal populations of this species by the means of a stepping-stone effect. To our knowledge, this is the first study to combine these two approaches to address connectivity among artificial offshore structures.

| Study area
The North Sea is located in the North-East Atlantic and is largely enclosed by land. Tidal water enters the North Sea in the south through the English Channel and from the north between Scotland and Norway. This results in a residual anticlockwise circulation. The North Sea bottom consists largely of sandy and muddy sediments, interrupted by areas with coarse substrates such as gravel and rocks.
Many anthropogenic structures provide artificial hard substrates in the North Sea. There are >27,000 shipwrecks, 1,397 offshore oil and gas production installations, ~3,500 wind turbines and a large (but not exactly known to us) number of navigational buoys ; Wind Europe, 2017).  Between 50 and ~100 individuals were sampled randomly at every location and stored at −20°C or in 70% ethanol to be transported to the laboratory. All samples were then cleaned of marine growth and placed at −80°C for long-term storage.

| Molecular methods
From each sample, between 24 and 67 specimens were selected randomly and genomic DNA was isolated from the adductor muscles by

| Connectivity calculated by particle tracking models
The transport of mussel larvae between sampled subpopulations in the North Sea was modelled using two Delft3D software modules: FLOW and PART (Deltares, 2016b(Deltares, , 2016a). The PART module was able to simulate midfield water quality and particle tracking, based on a hydrodynamic forcing output from the other Delft3D module, FLOW.

| Hydrodynamical model
The hydrodynamical model applied was Delft3D-FLOW, which solves the unsteady shallow-water equations in three dimensions.
The model incorporates a large number of processes, such as wind shear, wave forces, tidal forces, density-driven flows, stratification, atmospheric pressure changes, air temperature, and the exposure and inundation of intertidal flats. The large number of processes included in this module means that Delft3D-FLOW can be applied to a wide range of environments (e.g., river, estuarine, coastal and marine areas; Lesser, Roelvink, Kester, & Stelling, 2004). Flow equations were solved on a curvilinear grid consisting of 8,710 computational elements (Roelvink, Jeuken, van Holland, Aarninkhof, & Stam, 2001). The vertical resolution of the model was 10 water layers using a sigma-coordinated approach (i.e., proportional to water depth; Stelling & van Kester, 1994). Hydrodynamic transport was computed using detailed bathymetry and open boundary forcing based on tidal constituents. The model was forced using meteorological data from the High Resolution Limited Area Model (KNMI, 2015), which comprised two horizontal wind velocity components (at 10 m above mean sea level) and other atmospheric variables such as air pressure and temperature, archived every 6 hr. The freshwater discharges from 18 rivers were included in the model; seven of these discharges varied temporally (daily averages) and 11 were constant (based on long-term averages). This model is described in detail in Erftemeijer, Beek, Bolle, Dickey-Collas, and Los (2009

| Particle transport model
Particle tracking models are often used in environmental modelling (Broekhuizen, Lundquist, Hadfield, & Brown, 2011;North et al., 2008;Postma, Beek, Boogaard, & Stelling, 2013). Here, the Delft3D-PART module was used to calculate larval transport across the southern North Sea. Delft3D-PART is a random walk PTM, based on the principle that movement of dissolved substances in water can be described by a limited but potentially large number of discrete particles that are subject to advection due to currents and by horizontal and vertical dispersion. The movement of the particles in the model consists of two steps: advection, which is driven by the FLOW results; and dispersion, which is a stochastic random walk process.
In addition, the horizontal and vertical movement of the particles can be adjusted to account for swimming preferences or changes in buoyancy. Particle tracking allows water quality processes to be described in a detailed spatial pattern, resolving subgrid concentration distributions. Delft3D-PART is shown to be locally mass conservative (Postma et al., 2013). terns during the months we used for our PTM (i.e., March-June). The 2 years were especially comparable with regard to wind directions and wind force. We could only roughly assess this comparability from the meteorological data available to us (Royal Dutch Meteorological Institute); we did not perform a detailed analysis. We assume that through this approach the differences between the hydrodynamic circulation patterns in the (southern) North Sea between these two periods is less than when random years would have been used. For the Delft3D-PART model, each sampling location acted as an origin point of larvae, and the density of the larvae throughout the consecutive months at the other locations (destinations) was calculated.

| Modelling set up
Larval release was started on day 59 (1 March) of each year, from the top layer in the model. Both the timing and the period of particle release were based on the general reproductive timing and pelagic larval stage period of M. edulis. Mussels start spawning in early spring, based on temperature of the water. Variation of spawning timing may result from the variation in temperature increase between years; in the southern North Sea this usually starts in March, and peaks in April/May or even later (De Vooys, 1999). Time from spawning of mussel larvae to settlement is around 3 months (Widdows, 1991).
It was further assumed that the larvae are largely passive and that lis larvae is expected to be very high (Barker Jørgensen, 1981), this study did not include this aspect in the Delft3D-PART model. The purpose of the study was to assess the maximum success of larvae settling at a specific destination, which was assumed to be proportional to the total number of larvae arriving at a specific destination throughout the 70 days. The total number of arriving particles from each source-destination combination was calculated and combined in a particle matrix.

| Hybrid removal
A mix of alleles of M. galloprovincialis and M. trossulus was observed in our data at the partially species diagnostic locus Me15/16, and therefore we estimated the hybrid index h for each individual using a maximum likelihood approach that estimates the proportion of alleles inherited from one of two hybridizing species (Buerkle, 2005;Buerkle & Lexer, 2008). This approach was applied as implemented in the r package introgress (Gompert & Alex Buerkle, 2010). The samples from Lisbon (M. galloprovincialis) and Limfjorden (M. edulis) were used as parental reference samples and locus Med379 was not taken into account due to high levels of missing data for several samples.
All individuals identified as nonpure M. edulis (h < 1.0) were omitted from subsequent analyses, as this aspect would have affected the non-neutral influences on connectivity inferences.

| Population genetic statistics
Population-level analyses included pairwise Wright's F ST calculations (Wright, 1943(Wright, , 1965 and analysis of molecular variance (AMOVA), performed using genalex 6.5 (Peakall & Smouse, 2006, r (R Core  (Vavrek, 2011). In addition to these population-level analyses, the possible presence of genetic structure was also tested for by using Bayesian clustering algorithms implemented in the software structure version 2.3.4 (Pritchard, Stephens, & Donnelly, 2000). These algorithms are able to detect structure in the form of subgroups in the sampled individuals without a priori assuming that this structure is organized according to sampling location. With structure, no-admixture models were run with burn-in times of 100,000 and 100,000 repeats after burn-in with number of groups (K) ranging from 1 to 25.

| Isolation with migration analysis
Effective gene flow between sampling locations was estimated for all possible pairs, using the coalescent approach implemented in Correspondence between 2N e m values stemming from F ST estimation and from isolation-with-migration simulations was compared using Pearson's rank correlation. Correlation was performed with the average between the two (bidirectional) migration rates stemming from isolation-with-migration simulations.

| Migration and population differentiation versus arriving particles
To validate our hypothesis, we tested if the hydrography could partly determine connectivity of mussel populations growing on offshore structures in the southern North Sea. Thus, we tested the prediction that the number of arriving particles in the hydrogeographical model correlates with the genetic population differentiation statistic F ST as well as with migration rate as inferred from coalescent simulations.
The statistical significance of these correlations was assessed by a Mantel test (Mantel, 1967). Most implementations of the Mantel test assume that matrices whose correlation is to be tested are symmetrical; that is, when in a matrix M for all cells m i,j = m j,i where i equals row number and j equals column number. However, this is not generally true as tidal currents that are expected to drive connectivity differ in current speed between ebb and flow (McCave, 1971), and therefore in this work, the particle and migration rate matrices were asymmetrical. To determine the significance of the correlation between the isolation matrix and the particle matrix, we implemented the Mantel test ourselves in r. The Mantel test calculates the fraction of random permutations of the matrices that have a higher correlation compared with the original matrices. If the two matrices are unrelated, we expect half of the permutations to have a higher correlation (and the other half to a have a lower one). Here, correlation is the Pearson product-moment correlation coefficient. The number of permutations used was 10,000.

| RE SULTS
The number of alleles per locus varied between three and 69 (Table 3) with between 8% and 28% null alleles. The resulting data set had 23.14% null alleles. These high numbers of null alleles are in agreement with previously published data on microsatellite analyses in marine bivalves and gastropods (Bierne et al., 2003;Carlsson, 2008 in Mytilus spp. is pervasive and therefore this single locus was not sufficient for detecting a hybrid status of individuals, but a hybrid index proved to be more informative (Figure 3). Using Lisbon (N = 37) and Limfjorden samples (N = 48) as reference samples in the hybrid analysis, a wide range of hybrid indices was detected among the remaining individuals (Figures 3 and 4). Omitting all nonpure M. edulis  Figure S4). The analysis of possible population patterns without a priori classification (program structure) showed that there was no clear support for a particular number of subgroups within the data. This was indicated by the gradual increase of X|K (the likelihood of the data given number of groups K) with number of groups K ( Figure S5), which is here cautiously interpreted as either no support for obvious subgroups (Kalinowski, 2011;Porras-Hurtado et al., 2013), or, alternatively, that many small groups exist with subtle differences between them. Figure S6 shows structure bar plots for K = 2-7. Inferred migration per generation between sampling locations (   The Pearson product-moment correlation between pairwise F ST and arriving particles was −0.16 (p = .06), indicating that F ST was low between pairs with high particle exchange rates. The correlation between migration rate and arriving particles was .002 (p = .13). The effective travel distance and direction of particles thus explains part of the variation in pairwise F ST but not for migration rate.

TA B L E 3 Information per locus
Sample combinations without particle exchange had a 35% higher pairwise F ST compared to combinations with particle exchange (Figure 7a,b; p = .10, Kolmogorov-Smirnov test), being in line with the Mantel test. For the migration per generation between subpopulation combinations with and without particle exchange this difference was not apparent (Figure 7c,d; p = .44).

F I G U R E 5
Heat map showing migration rates among sampled locations. Estimated migration rate among all locations, red = high migration, green = low migration. x-axis: origin population, y-axis: destination population. Locations have been ordered in increasing distance to the nearest shore (

| D ISCUSS I ON
This study contributes with novel results to the understanding of the marine stepping-stone theory (Adams et al., 2014;Henry et al., 2018). Although PTM results predicted that Mytilus larvae should be unable to reach locations >85 km from coastal mussel populations, Mytilus spp. were found at all inspected locations, up to 181 km from the nearest coastline. Colonization of such far from shore locations by the mussels would require an adult population within a radius of 85 km. However, our analysis was unable to show that migration rates followed a pattern explained by modelled particle connectivity. Furthermore, although we found an indication that F ST was lower between connected locations, isolation by distance analysis showed no increase in isolation with increasing distance. Finally, we did not find clear structure among the populations. For example, many of the investigated locations were isolated from nearby locations.
Although mussels were present far offshore, we did not show them to originate from the intermediate stepping-stone locations investigated by us. The mussels colonized the available structures from unknown sources that placed larvae at offshore locations in an F I G U R E 6 Heat map showing particle exchange rates among sampled locations. Estimated particle exchange rate among all locations, red = high particle arrival, green = low particle arrival. x-axis: origin population, y-axis: destination population. Locations have been ordered in increasing distance to the nearest shore (  Although M. edulis can be observed in deeper waters (Table 1), it is most abundant in shallow waters as its presence is limited by predation in deeper waters (Langhamer, 2010). Consequently, large reproductive individuals of M. edulis are effectively absent on natural offshore reefs without intertidal zones .
The addition of many intertidal surfaces to the North Sea by artificial structures provides novel habitats for M. edulis (Krone et al., 2013).
Without these shallow hard substrates, M. edulis would be unlikely to survive in offshore North Sea waters.
At all locations in the North Sea, hybridization between M. edulis and M. galloprovincialis was observed. This extends the known hybrid zone between these taxa further north than hitherto described (Bierne et al., 2003;Fraïsse et al., 2014). Possibly, as earlier studies were restricted to nearshore locations, artificial offshore hard substrates seem to house more mussel hybrids than nearshore substrates. Alternatively, M. galloprovincialis may have spread northwards in recent years. The latter hypothesis awaits new nearshore genotyping of blue mussels, but the fact that our own nearshore sampling sites were found to contain many hybrids (Figure 3)  hydrodynamic regimes, we nevertheless detected a correlation between the pairwise F ST and arriving particle matrices.
Considering the young age of the artificial substrates we studied, and hence also of the populations of blue mussels living on them, the level of population genetic differentiation between many of the samples was remarkably high. On an evolutionary timescale these populations have been established extremely recently, as all offshore locations investigated were less than 50 years old. The range in population sizes must have been high, varying between an estimated <50,000 on a fully overgrown buoy, 50,000,000 on a medium-sized platform and billions on coastal locations, following generalized installation sizes and average M. edulis densities on offshore structures (Coolen et al., 2018;van der Stap, Coolen, & Lindeboom, 2016). Population genetic differentiation requires a number of generations on the order of the effective population size (the number of successfully reproducing individuals in a random mating population) to develop into an equilibrium state. The small initial population sizes may have accelerated the process of differentiation. We carried out coalescent analyses of isolation-with-migration that do not assume equilibrium between drift and migration (Hey & Nielsen, 2007;Sethuraman & Hey, 2016). The migration analysis, however, was found to not corroborate the correlation between genetic differentiation and hydrogeographical modelling. Founder effects (Nei, Maruyama, & Chakraborty, 1975) may have further obscured patterns in migration rates. If rare events have caused colonization of the structures from nearby but unconnected populations, the founder effect would cause the newly established population to seem isolated even though it originated from neighbouring populations. This founder effect is likely to increase after marine growth removal, as performed every 2-5 years on platforms and buoys (Coolen et al., 2018), which provides clean surfaces for new colonizers, resetting the age of the subpopulation every few years (Kerckhof & Cattrijsse, 2001).
It is possible that the pairwise population comparisons as modelled in ima2p (Hey & Nielsen, 2004;Sethuraman & Hey, 2016) did not fit well enough to the actual processes driving the situation in the field, causing uncorrelated results in our migration rates analysis. However, we did find a correlation between migration rates from coalescent analysis and rates based on F ST . This indicates that modelled migration rates fit the actual processes underlying genetic differentiation to some extent. Part of the colonization of offshore artificial hard substrates may have been the result of exceptional events (e.g., abnormal weather patterns, followed by isolation). Consequently, settlement on an offshore object by drifting mussel larvae might be rare. If such colonization events are rare and mussel stocks on offshore installations are isolated from each other most of the time, estimated migration rates would not correlate with total arriving particles as predicted from hydrogeographical models. This would be reflected in the observed patterns of genetic differentiation, underlining the rare colonization processes.
Blue mussels also occur on the hulls of ships and can be attached to flotsam (Apte, Holland, Godwin, & Gardner, 2000;Hopkins & Forrest, 2010;Thiel & Gutow, 2005). The introduction of larvae to offshore structures may therefore have been facilitated by anthropogenic activities. A potential source of larvae may have been jack-up rigs that are used near offshore installations such as platforms. These jack-ups can be covered in marine growth with high abundance of mussels (J.W.P. Coolen, personal observation). They are placed offshore near installations for various periods of time and are then relocated to other work sites offshore. Adult mussels on these jack-up rigs may spawn and cause colonization by mussels that are not closely related to nearby populations.

Mussels have been transported historically between the Wadden
Sea and waters near Breskens (Muehlbauer et al., 2014), providing higher genetic connectivity than would be expected from isolation by distance. The resulting genetic patterns may have obscured connectivity signals of the investigated subpopulations. This effect is supported by the low F ST for WZA and BRES, which are 380 km apart. These exchange activities should be addressed in future analyses, for example by following the methodology of Hernawan et al. (2017).
A further reason that may have led to the noncorrelation of estimated migration rates and modelled number of arriving particles could be that the particle arrival rates between sampling locations did not represent the actual hydrodynamic processes, which are underlying the exchange of larvae between these locations.
Accordingly, comparison studies revealed large variations between different PTMs (Hufnagl et al., 2016). Particle models are driven by hydrodynamic models that are driven by meteorological forcing, which may influence tidal movements and residual currents in the North Sea. Changing the meteorological forcing influences the outcome of the model. Furthermore, changing the time of spawning, length of spawning and survival period of larvae as modelled in the PTM is also likely to change the outcome of the model, because usually the weather changes over the season. Our choices are based on an "average" year, without predominant and strong easterly or westerly winds, and outflow from rivers, as well as average data on spawning and development of mussel larvae. Significant changes are therefore only expected in case of strong deviations in wind direction and force, and temperature in the spring. Because the collection of the mussels occurred over various years, during which the spring weather was mostly average, there is no reason to assume other input variables than chosen for the model runs.
Furthermore, many sample combinations were not connected according to the PTMs, resulting in many zeros in the arriving particles matrix. This was probably due to an average distance between locations that was larger than the average distance travelled by larvae. Future research should address this by sampling locations in closer proximity than in the current study. This will decrease the number of location combinations with zero particle exchange and is likely to better describe connectivity.
The stepping-stone effect possibly contributes to changes in distribution of nonindigenous species (Adams et al., 2014). Here, we used an indigenous species as model organism to test the concept of marine stepping-stones in the North Sea. In that region, nonindigenous species such as the Japanese amphipod Caprella mutica were found at the same offshore energy installations (Coolen et al., , 2018. The spread of native and non-native species depends largely on reproductive parameters and dispersal strategies. In the North Sea, any stepping-stone effect will be of most importance for epifaunal species with pelagic larval stages of several weeks (e.g., the oyster Ostrea edulis with 10-30 days or the limpet Crepidula fornicata with 14-21 days; Berghahn & Ruth, 2005;Shanks, 2009). Species with much shorter pelagic stages may be unlikely to reach the next subpopulation whereas species with very long pelagic larval stages may be capable of bridging large distances and may not need stepping-stones to colonize distant locations (Reisser, Bell, & Gardner, 2014). Other organisms, such as the nonindigenous Caprella mutica, also use floating objects such as ship hulls or debris for dispersal (Thiel & Gutow, 2005), and some species are transported by ballast water (Drake & Lodge, 2004), enabling them to colonize distant locations. This may be the most important factor driving the invasion by some non-native species .
The importance of artificial hard substrates as stepping-stones remains unclear. They may increase the speed of (future) bio-invasive events, as they provide valuable habitat in isolated locations for nonindigenous species. However, other factors such as hull fouling and ballast water exchange also strongly affect the distribution of nonindigenous species. For newly built installations, effects need to be evaluated in the context of tens of thousands of other artificial objects already present, including shipwrecks that have been there for more than 100 years .
In conclusion, artificial structures provide the only suitable habitats for species such as M. edulis at far offshore locations in the North Sea. Our results suggest that unexplained and rare colonization events may act together with average migration patterns to shape population genetic patterns and population connectivity.
Anthropogenic activities may have influenced genetic exchange between locations, for example via hull fouling on offshore vessels such as jack-up rigs. More detailed analyses of interannual variability in predicted larval transport patterns and inclusion of historical activities would be desirable. Future research should decrease the distance between modelled locations, to decrease the number of population combinations with zero connectivity in the modelled particle data.

DATA AVA I L A B I L I T Y S TAT E M E N T
All microsatellite data, migration per generation data and parti-