Historical human activities reshape evolutionary trajectories across both native and introduced ranges

Abstract The same vectors that introduce species to new ranges could move them among native populations, but how human‐mediated dispersal impacts native ranges has been difficult to address because human‐mediated dispersal and natural dispersal can simultaneously shape patterns of gene flow. Here, we disentangle human‐mediated dispersal from natural dispersal by exploiting a system where the primary vector was once extensive but has since ceased. From 10th to 19th Centuries, ships in the North Atlantic exchanged sediments dredged from the intertidal for ballast, which ended when seawater ballast tanks were adopted. We investigate genetic patterns from RADseq‐derived SNPs in the amphipod Corophium volutator (n = 121; 4,870 SNPs) and the annelid Hediste diversicolor (n = 78; 3,820 SNPs), which were introduced from Europe to North America, have limited natural dispersal capabilities, are abundant in intertidal sediments, but not commonly found in modern water ballast tanks. We detect similar levels of genetic subdivision among introduced North American populations and among native European populations. Phylogenetic networks and clustering analyses reveal population structure between sites, a high degree of phylogenetic reticulation within ranges, and phylogenetic splits between European and North American populations. These patterns are inconsistent with phylogeographic structure expected to arise from natural dispersal alone, suggesting human activity eroded ancestral phylogeographic structure between native populations, but was insufficient to overcome divergent processes between naturalized populations and their sources. Our results suggest human activity may alter species' evolutionary trajectories on a broad geographic scale via regional homogenization and global diversification, in some cases precluding historical inference from genetic data.


| INTRODUC TI ON
Humans are moving species beyond the limits of their natural dispersal abilities at unprecedented rates (Seebens et al., 2017). Humanmediated dispersal erodes biogeographic boundaries (Capinha, Essl, Seebens, Moser, & Pereira, 2015) and is expected to alter the evolutionary trajectories of species associated with human movement through both diversifying and homogenizing effects (Otto, 2018).
Whether human-mediated dispersal alters patterns of gene flow in native ranges is difficult to determine because the contributions of natural and human-mediated dispersal to gene flow are difficult to disentangle when both processes are ongoing. This is particularly problematic for species whose patterns of colonization and connectivity are understood primarily through genetic data, which is common for many introduced species. For example, the tunicate Ciona intestinalis exhibits genetic structure that is not predictable from either geography (as expected from natural dispersal) or shipping networks (as expected from human-mediated dispersal) among native populations in the English Channel (Hudson, Viard, Roby, & Rius, 2016), with debate over whether populations in North America result from human-mediated dispersal (Hudson, Johannesson, McQuaid, & Rius, 2019) or natural dispersal (Bouchemousse, Bishop, & Viard, 2016). In many cases, genetic patterns in native ranges are investigated only after the species has been introduced elsewhere, allowing vector activity to potentially influence native populations before they have been observed for the first time. The potential for human-mediated dispersal to alter patterns of gene flow is therefore likely underestimated, both in terms of the number of species it affects and its magnitude within species.
Here, we investigate how human activity reshapes evolutionary trajectories by exploiting a system of historical changes to shipping technology that resulted in a temporary period of human-mediated movement of intertidal sediments and the species associated with them. In the 10th Century, the adoption of wooden ships that used sediments shoveled from the intertidal and dredged from nearshore for ballast enabled long-distance exchange of abiotic materials and inadvertent hitchhikers in the European Atlantic (Jones, 1976;Stikeman 1832;McGrail, 1989;Ansorge, Frenzel, & Thomas, 2011) and Mediterranean (Casson, 1995). During the 11th-15th Centuries, the magnitude and reach of these practices increased with the rise of maritime trade around the North and Baltic Seas, with annual rates of sediment relocation in some estuaries on the order of millions of tons (McGrail, 1989;Stikeman, 1832). The exchange of sediments between estuaries via ballast extended in geographic reach during the mid-15th Century with European exploration and trade in North America. This exchange abruptly declined, and indeed likely ceased entirely, in the late 19th Century when wooden vessels were replaced by steel-hulled ships that used seawater as ballast (Fofonoff, Ruiz, Steves, & Carlton, 2003). This provides an opportunity to investigate how human-mediated dispersal impacts native populations: If historical human-mediated dispersal in the native range was weak, patterns of genetic differentiation and phylogeographic structure among these populations are expected to be more prominent than among populations in the introduced range, whereas homogenization by widespread human-mediated dispersal is expected to create similar patterns of genetic differentiation and population structure in both the introduced and native ranges.
We focus on two intertidal invertebrates, the amphipod Corophium volutator (Pallas, 1766) and annelid Hediste diversicolor (Müller, 1776, previously Nereis diversicolor), which inhabit soft-sediment habitat in the Bay of Biscay, North, and Baltic Seas in Europe and the Gulf of Maine and Bay of Fundy in North America. Until recently, they were thought to be native in North America (for review, see Einfeldt & Addison, 2015;Einfeldt, Doucet, & Addison, 2014).
Three main lines of evidence suggest they were once moved between Europe and North America in ballast sediments before sediment movement was reduced when water ballast was adopted: (a) these soft-sediment dwelling intertidal invertebrates are abundant in Europe and North America, but are not reported in intermediate land masses (Iceland and Greenland) which are expected to be inhabited by natural colonizers (Ingolfsson, 1992); (b) genetic diversity in both species is consistent with North American populations representing a subsample from more diverse European populations (Einfeldt & Addison, 2015;Einfeldt et al., 2014); (c) they lack a pelagic phase, brood-rear their young, and reside in shallow constructed burrows (Peer, Linkletter, & Hicklin, 1986;Scaps, 2002). This is expected to impart limited natural long-distance dispersal capabilities (via drifting, swimming, or oceanic rafting), as well as a strong association with sediments used in historical ballast, and a weak association with water ballast (Hewitt, Gollasch, & Minchin, 2009).
Neither species was found in surveys of modern ballast tanks from 62 ships arriving in North America (Briski, Ghabooli, Bailey, & MacIsaac, 2012), and only a single ship was found to carry any C. volutator specimens in a survey of 550 ships in Europe (Gollasch et al., 2002). Prior to their introduction to North America, the same vector that introduced these species from Europe to North America is likely to also have facilitated gene flow among native populations in Europe.
Trace fossils suggest that both species were present on the continental coast of the North Sea and British Isles during this time (Allen & Haslett, 2002;Buller & McManus, 1972;Streif, 1972). In the absence of human-mediated dispersal, it is therefore expected that population structure (reflected by genetic differentiation) and phylogeographic structure (reflected by geographically structured tree-like phylogenies) in C. volutator and H. diversicolor will reflect these physical barriers to movement.
To investigate the net effect of diversifying and homogenizing aspects of human-mediated dispersal on species' evolutionary trajectories, we compare patterns of genetic divergence using re-
In addition to the 78 H. diversicolor samples, we identified samples from the Mediterranean (n = 4), the North Sea (n = 5), and North America outside of the Bay of Fundy (n = 17) as cryptic species related to H. diversicolor from mitochondrial DNA (Einfeldt et al., 2014;Virgilio et al., 2009). Because we have not sampled the entire native range of these cryptic lineages of the H. diversicolor species complex originating in the Mediterranean and Ponto-Caspian Seas, they were excluded from further analyses. We extensively sampled soft-sediment habitat along the coast of the Northwest Atlantic from Québec to Massachusetts, and the populations presented here represent the entire present North American latitudinal range of both species.
To maximize the number of nucleotides kept with all sequences a standard length, we trimmed reads to 85 bp. Due to the lack of available genomic resources for C. volutator and H. diversicolor, to reduce nontarget sequencing we filtered reads for contaminants against standard human and microbial contaminant databases, and custom databases built from publicly available genomes for parasites and algae using Kraken v.1.1.1 (Table S3; Wood & Salzberg, 2014). We assembled a de novo reference and called SNPs using STACKS v.1.35 (Catchen, Hohenlohe, Bassham, Amores, & Cresko, 2013), with forward and reverse reads assembled as different stacks and requiring at least five identical reads for each stack (m = 5) to reduce errors introduced through sequencing. We compared results for several combinations of nucleotide mismatches per individual allowed to group stacks (M = 2, 4, 6, or 8) and nucleotide mismatches allowed among stacks to form catalog loci (n = 2, 4, 6, or 8), which showed a plateau in number of loci retained at M = 6 and n = 6. To reduce physical linkage, we retained only a single SNP per stack, and to reduce bias from missing data, we retained loci genotyped in 100% of populations (p = 1) and at least 50% of individuals per population (r = .5) and filtered rare variants by removing SNPs with minor allele frequencies <0.01 (Linck & Battey, 2019).

| Genomic diversity and structure
Evolutionary change between introduced populations and their sources can be facilitated by the introduction of genetic diversity from multiple sources (Brawley et al., 2009;Roman & Darling, 2007;Viard, David, & Darling, 2016) and founding effects that facilitate the random or selective establishment of some genotypes over others during the introduction process (Dlugosch & Parker, 2008;Wares, 2005). To assess founding effects associated with introduction, we calculated observed and expected heterozygosity using Adegenet (Jombart & Ahmed, 2011

| Simulated demographic scenarios
To investigate the impact of human-mediated migration on phylogeographic patterns, we conducted simulations following a complex demographic history with varying degrees of natural and human-medi- year, reflecting reproductive cycles of C. volutator and H. diversicolor (Dales, 1950;Peer et al., 1986). We set the magnitude of within-range human-mediated migration (M H ) active in this system over 1,800 generations (corresponding to the ~900-year window of widespread ballast sediment exchange) to be an order of magnitude less than, equal to, and an order of magnitude more than the magnitude of natural migration (M N ), followed by a return to natural migration matrices for 200 generations (~100 years).

| Demographic complexity and phylogenetic conflict
Species with limited capacities for natural dispersal are expected to exhibit tree-like phylogenetic patterns reflecting demographic history and geographic barriers to gene flow, while gene flow enabled by human-mediated dispersal is expected to decrease the phylogenetic concordance across loci, creating phylogenetic conflict reflected by low bootstrap support for nodes and high measures of reticulation.
To examine phylogenetic relationships in C. volutator, H. diversicolor, and simulated datasets, we created unrooted networks based on uncorrelated p-distances between individuals using a neighbor-net method in Splitstree v.4.13 with 1,000 bootstrap iterations (Huson & Bryant, 2005). To compare levels of phylogenetic conflict in empirical and simulated data, we quantified network reticulation by calculating the mean Delta scores across networks (Holland, Huber, Dress, & Moulton, 2002). High levels of Delta suggest phylogenetic conflict. Delta is a ratio of path lengths between quartets of populations, for which a value of 0 corresponds to low phylogenetic conflict and a completely tree-like phylogeny, and a value of 1 corresponds to equal distances between all populations in the quartet.

| Scans for selection
We performed scans for positive selection using two methods that are reported to perform well in demographic scenarios involving hierarchical spatial structure and population expansion: an F ST based approach implemented in OutFLANK v.0.2 (Whitlock & Lotterhos, 2015) and a principle components based approach implemented in PCAdapt v 4.1 (Luu, Bazin, & Blum, 2017). To correct for bias arising from multiple tests, we calculated q-values from p-values produced by each test with a false discovery rate threshold of 5%. For PCAdapt, we chose the optimal number of genetic components K based on Cattell's rule applied to scree plots, which was K = 3 for both species. To assess false-positive ratios, we performed both scans for selection on SNPs from data simulated using a strictly neutral model of evolution.

F I G U R E 2
Population genetic structure. a-b, Neighbor-joining networks for Corophium volutator (4,870 SNPs) and Hediste diversicolor (3,820 SNPs) show clustering of individuals by sampling site, division between native versus introduced ranges, and reticulation between populations within each range. Lighter network shading indicates weaker bootstrap support due to phylogenetic conflict. * Introduced samples. c and d, Admixture proportions for individuals in K optimal genotypic clusters with SNMF

| Samples and SNPs
We genotyped 121 C. volutator and 78 H. diversicolor individuals from mudflats covering the geographic extent of their distributions, with each sample site representing a population unit (Figure 1).
Sequencing 2 × 100 base-pair paired-end reads of double-digest restriction-site-associated DNA at >10× average coverage and keeping the first single nucleotide variation (SNP) per locus produced a total of 4,870 SNPs for C. volutator and 3,820 SNPs for H. diversicolor (Table S4).

| Reduced genetic diversity in North American populations
The number of SNPs polymorphic in one continent but not the other (corrected for the number of individuals) was 6.4 in North America   (Table S1). The proportion of genetic variation between continents was greater than the proportion of variation between populations within continents for C. volutator (14.7% between continents, 10.1% between populations within continents), but not for H. diversicolor (8.8% between continents, 17.5% between populations within continents). The proportion of variation between individuals within populations was relatively low for both C. volutator (5.2%) and H. diversicolor (2.3%).

| Genetic divergence between North America and Europe
In C. volutator, the best-fit number of genetic clusters (K = 2) differentiates North American from European individuals, and the best-fit value for each continent separately is K = 1 for North America and K = 2 for Europe, with two populations from Ireland (ROS & BAL) and one from Denmark (BAH) forming a separate cluster (consistent with results for Figure S1 K = 3). Each of these clusters analyzed separately has a best-fit value of K = 1. In H. diversicolor, the best-fit number of clusters (K = 4) differentiates North American individuals from three genetic clusters in Europe, and analyses of these clusters separately have a best-fit value of K = 1. Increasing K beyond its optimal value reveals increasingly fine-scale clustering that corresponds to geography in both species, particularly among native populations, but does not group North American individuals with European clusters ( Figure S1).

| Genetic structure among European populations
Phylogenetic networks and genetic clustering recover nearly all population groups with high fidelity (Figure 2, Figure S1), consistent with a scenario of limited natural dispersal and fragmented habitat.

| Phylogenetic conflict between European populations
Long-standing isolation between continental Europe and the British Isles and stepping-stone migration between populations distributed along >2,500 km of European coast is expected to produce branching networks that reflect colonization history, geographic location, and major barriers to dispersal (

| Human-mediated gene flow erodes ancestral phylogeographic structure
To

TA B L E 2 Phylogenetic conflict in
Corophium volutator, Hediste diversicolor, and simulated data continental mainland begin to re-order into positions corresponding to their geographic location, and Delta further decreases to 0.29, reflecting a further decrease in mean phylogenetic conflict.

| High false-positive rates in scans for selection
The erosion of ancestral phylogeographic structure precludes accurate inference of premixture demographic history using bi-allelic SNP data (Appendix S1), but whether it violates the assumptions of selection tests relying on SNP frequencies is less obvious. To determine whether selection can be inferred from distribution-wide patterns of genetic diversity in this system, we performed scans for positive selection on C. volutator, H. diversicolor, and data from simulations using two methods reported to perform well regardless of demographic history: OutFLANK (Whitlock & Lotterhos, 2015), which infers the neutral distribution of F ST , and the PCA-based PCAdapt (Luu et al., 2017). OutFLANK did not detect outlier SNPs in either species or the simulated data sets. PCAdapt identified a large proportion of SNPs as outliers in C. volutator (16% of SNPs) and H. diversicolor (23% of SNPs), but these results were well within the 13%-46% (mean α = .27, SD = 0.11) range of false-positive ratios identified in simulated data (Table S2). These results suggest that the demographic histories of these species shifted allele frequencies strongly enough to mask differences resulting from directional selection even if it is present, precluding accurate inference of selection F I G U R E 3 Human-mediated migration erases ancestral phylogeographic structure. a, Demographic model for simulated genetic data: population expansion, post-Pleistocene colonization from a glacial refugium, natural migration (M N ) between adjacent populations in discontinuous habitat, founding effects during introduction, and temporary human-mediated migration (M H ) based on two generations per year. * N e reduced by 50% in populations 7 and 15, by 90% in population 16 relative to majority of populations to visualize the effects of increased drift. b, Neighbor networks of SNPs from simulated genetic data for different values of M N and M H , with the lowest values in the top left representing a scenario with the least amount of dispersal. c, Time series of a single simulation (M N = 5 -4 , M H = 5 -4 ) shows initial phylogeographic structure, erosion of branches after the onset of human-mediated dispersal, and early stages of phylogeographic structure 100 years after human-mediated dispersal has ceased using frequency-based scans. Whether selection contributes to the patterns of genetic divergence observed from genome-wide SNPs in C. volutator and H. diversicolor thus remains uncertain.

| D ISCUSS I ON
We exploited historical introductions of two intertidal invertebrates with limited natural capabilities for dispersal to determine the effects of human activity on their evolutionary trajectories. North American populations had lower metrics of genetic diversity than European populations, consistent with both species having undergone founder effects or genetic bottlenecks as a result of introduction from Europe, which may contribute to divergence between the introduced and native ranges. Native populations are expected to exhibit phylogeographic structure reflecting landscape features and demographic history, and if structure is present in a native range then introduced populations are expected to be more genetically similar to their sources than other native populations (Geller, Darling, & Carlton, 2010). However, we found genetic divergence between the introduced and native ranges of both species and extensive phylogenetic conflict among populations within either range. These results support the hypotheses that (a) human-mediated dispersal can facilitate establishment of introduced populations that follow evolutionary trajectories that are independent from their native sources if human activity ceases; and (b) human activity can overcome natural barriers to dispersal between native populations and erode ancestral phylogeographic structure.
We detected evidence of contemporary gene flow in C. volutator or H. diversicolor between populations in adjacent mudflats in a limited number of cases, suggesting that rates of natural dispersal vary by location due to interactions with environmental features such as regional currents. Natural dispersal of these species over short distances within either continental coastline is likely facilitated by brief bouts of swimming associated with incoming tides (Aberson, Bolam, & Hughes, 2011;Drolet & Barbeau, 2009)  However, DNA from individuals that do not group with the expected genetic clusters (LAV01, LAV02, BAN02, and BAN06) was adjacent on plates during library preparation, and mislabeling cannot be ruled out as a potential cause of this pattern. If these two estuaries are connected, the likely recent connectivity between them appears to be a rare exception to the general pattern of contemporary isolation between sites. While modern ballast water transport is a contemporary dispersal vector for other crustaceans and annelids (Briski et al., 2012), the limited evidence of contemporary connectivity be- species, impeding inference of historical colonization routes (further discussed in Appendix S1). This contrasts with genetic patterns from mitochondrial DNA, which found North American haplotypes to be subsampled from those found in European populations with no discernible divergence (Einfeldt & Addison, 2015;Einfeldt et al., 2014).
Noncongruent patterns between mitochondrial and genomic DNA are expected due to recombination between nuclear markers but not mitochondrial sequences, consistent with patterns seen in more recently introduced species (Ayari et al., 2017; Jeffery et al., 2017).
Divergence between Europe and North America may be in part due to rapid evolution of introduced populations, which can arise from neutral (e.g., Baker & Moeed, 1987;Shultz, Baker, Hill, Nolan, & Edwards, 2016) and selective (e.g., Colautti & Barrett, 2013;Huey, Gilchrist, Carlson, Berrigan, & Serra, 2000) processes and ultimately lead to reproductive isolation between introduced populations and their sources (e.g., Montesinos et al., 2012). We detected few variants unique to the introduced range of either species (Table 1) Bohonak, 1999). Populations evolving in isolation are expected to diverge in monophyly, with IBD resulting in phylogenetic patterns that reflect geography, while gene flow between populations is expected to cause conflicting phylogeographic signals between genomic regions (Avise, 2000;Maggs et al., 2008). We expected greater pairwise genetic differentiation and stronger patterns of IBD in Europe than in North America due to greater geographic separation and longer time since becoming naturalized. IBD was evident in Europe and North America for both C. volutator and H. diversicolor, but counter to our predictions it was stronger in North America, suggesting that the factors restricting movement differ between ranges or over different geographic scales. Long-standing IBD resulting from natural dispersal between adjacent populations is expected to produce phylogenetic networks with topologies that reflect geography (e.g., Jeffries et al., 2016), and limited dispersal between mainland and island habitats is expected to produce branching networks (e.g., first network in Figure 3c). Contrary to these expectations, The potential for human activity to facilitate genetic admixture between native populations may have both undesirable and desirable consequences for conservation. Human-mediated dispersal may increase the risk of species to become invasive beyond their natural distributions by facilitating secondary contact among native populations and thereby increasing their adaptive potential (e.g., Anderson et al., 2018;Bertelsmeier et al., 2018;Lehnert et al., 2018;Vandepitte et al., 2014). We provide the first empirical evidence that human-mediated dispersal may decrease our ability to interpret history from genetic data in native as well as introduced ranges by obscuring ancestral phylogeographic structure arising from natural processes, erasing the signatures of demographic history recorded in DNA (Crozier & Schulte-Hostedde, 2015). On the other side of the same coin, human-mediated gene flow could enhance the introduction of advantageous alleles to populations that are not otherwise able to respond to anthropogenic pressures such as climate change, akin to directed efforts of assisted gene flow (Aitken & Bemmels, 2016).
The effect of human-mediated movement on patterns of gene flow within native ranges is likely underestimated. The native ranges of other introduced marine species often exhibit unexpected genetic patterns, which could potentially be explained by human patterns of human movement. For example, in the invasive green crab Carcinus maenas mitochondrial haplotypes typically found in the South of Spain have also been found in the UK (Darling et al., 2008), and an The degree of association with contemporary vectors of transport is unknown for many taxa, due to limited availability of vessel surveys and the challenges of correctly identifying small inconspicuous species across all their life stages (Haydar, 2012). Additional genetic analyses of historically introduced species could help further our understanding of how frequently each of these unintentional outcomes has occurred (Rius & Darling, 2014), providing insight into expected evolutionary outcomes of contemporary human activity.

ACK N OWLED G M ENTS
We acknowledge NSERC for funding provided to JAA, ACEnet and

CO N FLI C T O F I NTE R E S T
The authors declare that they have no potential sources of conflict of interest.