Hemigrapsus sanguineus were collected by hand and with traps on the island of Helgoland, Germany, between the 2nd and 4th of August 2017. Crabs were collected during the day at low and incoming tide. The traps were placed in the subtidal zone at approx. 0.3–0.5 m depth for 1 h, meanwhile crabs were collected by turning rocks in the intertidal zone. The traps were constructed out of PVC pipe (approx. 500 mm long and 100 mm Ø) with a funnel (25 mm Ø opening) in one end and a 1 mm steel mesh in the other end. In the middle of the PVC pipe a drill out cylindrical tube was fitted, containing a crushed Mytilus edulis. The traps were weighted down by two heavy chains, attached to each side of the trap (total weight of approx. 5 kg).
The crabs were then brought to the Sven Lovén Centre for Marine Infrastructure—Kristineberg, Sweden by transport in an electrical cooling bag, set to 10 °C to minimize activity and potential cannibalism, filled up with Fucus spp. The transportation took approx. 48 h, no crabs died during transit.
Individuals of C. maenas were collected by hand and with the use of traps, in Smalsund, outside of Kristineberg on the 9th of August 2017. The traps were of the same design as those used in Helgoland. Traps were deployed in the subtidal zone at a depth of 0.2–0.5 m and left them for approx. 1 h, during which crabs were collected by turning rocks in the intertidal.
The crabs were kept in separate holding tanks before and during the experiments. The water in the tanks was filtrated sea water with a salinity of 25 ± 1 PSU and a temperature of 19 ± 1 °C. At the time of collection, the waters in Helgoland had a salinity of 33 PSU and a temperature of 19 °C.
Competition experiments with Carcinus maenas
Experiments started on the 15th of August, 11 days after the last sampling day for H. sanguineus and 6 days after the sampling of C. maenas. The experiments comprised three stages (early invasion, mid invasion, late invasion) always containing 10 crabs per aquaria but with different proportions of species and sex depending on the stage (Table 1). These stages are to reflect on the different species ratios during a potential invasion: invader species in minority, in equal number to the native, and eventually in majority. The sex ratio for both is based on the sex ratio of C. maenas found in the field along the Swedish west coast (Jungblut and Karlsson unpubl. data).
Crabs used in the experiments had both chelae intact and at least 2 legs on each side of the body, no newly moulded crabs were used. The carapace width (CW) of crabs used in the experiments had the same dimensions in both species: 20–35 mm CW for males, and 15–25 mm CW for females. To distinguish males and females apart during the experiment, red nail-polish was applied to the carapace of the females. Crabs were randomly collected from the holding tanks. Due to limited number of H. sanguineus, crabs were returned to the holding tank after the experiment. Hence these crabs may have been used multiple times. However, after each round of experiments, all crabs were feed M. edulis and then rested for at least 48 h until the next experiment started.
Three experimental aquaria were recorded simultaneously using a Logitech C920 webcam and the recording software OBS 20.0.1. The aquaria used had a surface area of 0.1 m2 (210 × 475 mm), resulting in a density of ~ 100 crabs/m2, a density common in the field. In each aquarium a thin layer of filtered shell-sand (1–5 mm grain size) without organic matter was used as a substrate.
The first experiment assessed if H. sanguineus can actively dislodge C. maenas and/or restrain its food sources at any of the three invasive stages. To this end, all crabs were kept without food in holding tanks with filtered seawater for 48 h. Thereafter, the animals were transferred to the experimental aquaria and a M. edulis (length 80 ± 10 mm with one side crushed) was placed in each aquarium. The aquaria were then recorded for 30 min. At 1, 5, 10, 15, 20, 25 and 30 min into the recording, presence or absence of feeding on the mussel was noted for each crab species and each of its sexes.
The mussels were then removed, and a man-made shelter was placed in each aquarium. The shelter was constructed by a PVC pipe (150 mm in length and 67 mm in diameter) which was attached to a plastic slab (200 × 125 mm) to keep the pipe in place. The shelter was then covered in a layer of sand and the crabs were then left for 30 min to settle down after being disturbed and then recorded for 30 min. At 1, 5, 10, 15, 20, 25 and 30 min into the recording presence or absence of crabs in the shelter was noted for each species and each sex. A minimum of half of the crab needed to be covered by the shelter to be considered “sheltered”. After the end of the experiment, the crabs were relocated to their holding tanks.
For each group (e.g. C. maenas males), a mean feeding ratio/hiding ratio was calculated based on the presence (1) or absence (0) at the food location/inside the man-made shelter at each time point (1, 5, 10 min etc.). This was done separately for each of the 6 replicates (aquaria). These 6 replicates were used to calculate the 95% confidence interval as well as a pairwise t test assuming unequal variance (Table S1).
Species and environmental data
The selected geographical area for this study include the coastal waters of Europe from the French west coast in the south to the Faroe Islands in the north, and also the Baltic Sea. Occurrence records (presence only) were obtained by sampling the Swedish West coast, through the literature (Dauvin et al. 2009; Dauvin 2009b; Van den Brink et al. 2012) as well as from the Global Biodiversity Information Facility (GBIF 2017a, b). Records obtained by the authors and collaborators were submitted to GBIF through the Swedish Species Gateway (http://www.Artportalen.se). All occurrence records were carefully revised and verified before being used in the models, while unverified records were discarded.
We used gridded environmental data available as global marine layers through Bio-Oracle (http://www.bio-oracle.ugent.be/) with a resolution of 5 arc-min (Tyberghein et al. 2012). These data layers are generated from monthly satellite data (Aqua-MODIS and SeaWiFS; https://oceancolor.gsfc.nasa.gov/) as well as in situ measured oceanographic data from the World Ocean Database 2009 (Boyer et al. 2009). We also used marine layers from AquaMaps (http://www.aquamaps.org/download/main.php) with a resolution of 30 arc-min (Kaschner et al. 2008). These layers were built from long-term averages of temporally varying environmental variables (Ready et al. 2010).
Environmental data was obtained using version 2 of the BioClim workflow (http://purl.ox.ac.uk/workflow/myexp-3725.2) available at the BioVeL portal (Hardisty et al. 2016). The workflow was used to retrieve environmentally unique points (Nix 1986; Vestbo et al. 2018) from the species occurrence files for a set of 14 environmental data sets (abbreviations used throughout the article are given in parentheses), including the following. Bio-Oracle (5 arc-min): Mean dissolved oxygen in ml/l (Oxy), Mean nitrate [NO3] [NO3 + NO2] in µmol/l (NO3), Mean phosphate in µmol/l (PO4), Maximum sea surface temperature in °C (Max SST), Minimum sea surface temperature in °C (Min SST), Sea surface temperature range in °C (Range SST), Mean calcite concentration in mol/m3 (CaCO3), Maximum chlorophyll A concentration in mg/m3 (Max ChlA), Minimum chlorophyll A concentration in mg/m3 (Min ChlA), Range of chlorophyll A concentration in mg/m3 (Range ChlA). AquaMaps (30 arc-min): Mean ice concentration in % (Ice), Mean sea surface salinity in PSU (SSS), Mean sea surface temperature °C (SST), Mean chlorophyll A concentration in mg/m3 (Mean ChlA). Chlorophyll A data sets were included as chlorophyll acts as a proxy for phytoplankton, and thus reflects the amount of nutrition in the water.
Initially, we conducted a correlation analysis (data not shown) as well as a principal component analysis (PCA) on log-transformed data using the R statistical environment 3.0.2 (R Core Team 2013). Both analyses were performed to identify the variables explaining the variation in the data set, and which can be used as predictor variables in the modelling. The PCA was also used to compare the environmental space occupied by the two species and estimated as a 9 and 10-dimensional hypervolume (Fig. S1). We used all non-correlated variables that were represented (with > 5%) in the first two components of the PCA to build the niche models.
Ecological niche modelling
We used version 20 of the ecological niche modelling (ENM) workflow (http://purl.ox.ac.uk/workflow/myexp-3355.20) to describe and compare the geographical and environmental space occupied by the two species and to estimate the potential distribution maps (PD) with favourable biotic and environmental conditions in the study region. We did not consider demographic, dispersal, or substrate properties that may also be used in species distribution modelling (Peterson et al. 2011; Reiss et al. 2014). For more information on the ENM workflows see De Giovanni et al. (2015), Holl et al. (2013), Leidenberger et al. (2015) and Vestbo et al. (2018). We executed parallel analyses with two ENM algorithms, Maximum Entropy v. 1.0 (Phillips et al. 2006; Phillips and Dudík 2008), and Support vector machine v. 0.5 (Schölkopf et al. 2001).
Models were created using each species’ maximum distribution range within the study region and the set of predictor variables identified in the PCA. Maxent models were set to run with 10.000 background points (including input points) drawn from the mask. Feature selection was automated, allowing the algorithm to combine feature types when fitting a model, and perform 500 iterations. Tolerance for detecting model convergence was set to 0.00001, while sample threshold was set to 80 (product), 10 (quadratic), and 15 (hinge). SVM models were set to execute the C-SVC algorithm with radial basis kernels, gamma values 1/k (where k is the number of layers), and a cost value of 1. All models were set to produce a probabilistic, instead of binary output.
For each species, we ran both algorithms across the above-specified environmental layers. Models were tested using fivefold cross-validation based on the area under the curve (AUC) value and omission error rate (false negative rate, OE), and subsequently projected using the same layers (native projections). The results of the ENMs were visualized as potential distribution maps (PD) maps, showing the suitable environment of a species in the region of interest. Overall, we executed 4 niche models (2 algorithms, 2 species).
Raster maps created by the niche modelling algorithms were processed using the qGIS software package v. 2.6 Brighton (Quantum GIS Development Team 2014). We produce PD maps as consensus from the raster values of both algorithms, where red indicates agreement between both algorithms according to lowest presence threshold in each cell (Pearson et al. 2007).