Population genomics of the introduced and cultivated Pacific kelp Undaria pinnatifida: Marinas—not farms—drive regional connectivity and establishment in natural rocky reefs

Abstract Ports and farms are well‐known primary introduction hot spots for marine non‐indigenous species (NIS). The extent to which these anthropogenic habitats are sustainable sources of propagules and influence the evolution of NIS in natural habitats was examined in the edible seaweed Undaria pinnatifida, native to Asia and introduced to Europe in the 1970s. Following its deliberate introduction 40 years ago along the French coast of the English Channel, this kelp is now found in three contrasting habitat types: farms, marinas and natural rocky reefs. In the light of the continuous spread of this NIS, it is imperative to better understand the processes behind its sustainable establishment in the wild. In addition, developing effective management plans to curtail the spread of U. pinnatifida requires determining how the three types of populations interact with one another. In addition to an analysis using microsatellite markers, we developed, for the first time in a kelp, a ddRAD‐sequencing technique to genotype 738 individuals sampled in 11 rocky reefs, 12 marinas, and two farms located along ca. 1,000 km of coastline. As expected, the RAD‐seq panel showed more power than the microsatellite panel for identifying fine‐grained patterns. However, both panels demonstrated habitat‐specific properties of the study populations. In particular, farms displayed very low genetic diversity and no inbreeding conversely to populations in marinas and natural rocky reefs. In addition, strong, but chaotic regional genetic structure, was revealed, consistent with human‐mediated dispersal (e.g., leisure boating). We also uncovered a tight relationship between populations in rocky reefs and those in nearby marinas, but not with nearby farms, suggesting spillover from marinas into the wild. At last, a temporal survey spanning 20 generations showed that wild populations are now self‐sustaining, albeit there was no evidence for local adaptation to any of the three habitats. These findings highlight that limiting the spread of U. pinnatifida requires efficient management policies that also target marinas.


Sample
Habitat Locality ( Table S3. Hierarchical analysis of the molecular variance (AMOVA) performed on the RAD-seq panel to examine the spatial (habitat type) and temporal effects on genetic structure. Samples from 2015, 3 groups (3 groups: marinas, farms, natural

Marinas (2 localities) grouped by year (3 groups: 2005, 2009, 2015)
Among groups -6.17 Each individual is represented by a vertical line divided into K coloured segments, the length of which indicates the individual's membership fraction to each of K clusters. Individuals are grouped according to their sampling locality (ordered along a south to north gradient) for the regional scale analysis, and according to locality and year of sampling for the analysis at the bay scale. Locality codes correspond to those specified in Table 1. Note that conversely to fastSTRUCTURE (see figures in the main text), the snmf algorithm does not rely on Hardy-Weinberg equilibrium assumptions, and is particularly appropriate to use with inbred species (Frichot, Mathieu, Trouillon, Bouchard, & Francois (2014). Fast and efficient estimation of individual ancestry coefficients. Genetics,196(4),[973][974][975][976][977][978][979][980][981][982][983]. The first two discriminant functions are displayed in the top plot and the second and third functions are displayed in the bottom plot. The sample codes, shown in the legend in the bottom right corner of each panel, correspond to those given in Table 1, and the symbols (shape and colour) refer to those displayed in Fig. 1

, C and D) using STRUCTURE (A and C) or INSTRUCT (B and D).
Each individual is represented by a vertical line divided into K coloured segments, the length of which indicates the individual's membership fraction to each of K clusters. Individuals are grouped according to their sampling locality (ordered along a south to north gradient) for the regional scale analysis, and according to locality and year of sampling for the analysis at the bay scale. Locality codes correspond to those specified in Table 1. STRUCTURE relies on Hardy-Weinberg equilibrium assumptions, unlikely to be met in a selfing species. Conversely, INSTRUCT does not rely on this assumption and jointly estimates ancestry coefficient and selfing rates. Note that OutFLANK is not included on the diagrams because no single outlier was detected at the 5% qvalue threshold.
A) Sample set 1 B) Sample set 2 Figure S5. Bayesian clustering analyses with the fastStructure software using alternative RAD-seq dataset RAD-seq panel with sample set 1 (marinas and natural sites sampled in 2015) used for outlier detection, with analyses done using A) 64 loci identified as positive selection outliers by OutFLANK with a q-value of 0.056 (see text), and B) the putative "neutral" loci, i.e. all SNP loci except the 240 outlier loci detected by at least one of the five methods used (including OutFLANK at the 5.6% threshold). OutFLANK is the most conservative outlier detection method.
Each individual is represented by a vertical line divided into K coloured segments, the length of which indicates the individual's membership fraction to each of K clusters. Individuals are grouped according to their sampling locality (ordered along a south to north gradient).