Genetic signals of artificial and natural dispersal linked to colonization of South America by non‐native Chinook salmon (Oncorhynchus tshawytscha)

Abstract Genetics data have provided unprecedented insights into evolutionary aspects of colonization by non‐native populations. Yet, our understanding of how artificial (human‐mediated) and natural dispersal pathways of non‐native individuals influence genetic metrics, evolution of genetic structure, and admixture remains elusive. We capitalize on the widespread colonization of Chinook salmon Oncorhynchus tshawytscha in South America, mediated by both dispersal pathways, to address these issues using data from a panel of polymorphic SNPs. First, genetic diversity and the number of effective breeders (N b) were higher among artificial than natural populations. Contemporary gene flow was common between adjacent artificial and natural and adjacent natural populations, but uncommon between geographically distant populations. Second, genetic structure revealed four distinct clusters throughout the Chinook salmon distributional range with varying levels of genetic connectivity. Isolation by distance resulted from weak differentiation between adjacent artificial and natural and between natural populations, with strong differentiation between distant Pacific Ocean and Atlantic Ocean populations, which experienced strong genetic drift. Third, genetic mixture analyses revealed the presence of at least six donor geographic regions from North America, some of which likely hybridized as a result of multiple introductions. Relative propagule pressure or the proportion of Chinook salmon propagules introduced from various geographic regions according to government records significantly influenced genetic mixtures for two of three artificial populations. Our findings support a model of colonization in which high‐diversity artificial populations established first; some of these populations exhibited significant admixture resulting from propagule pressure. Low‐diversity natural populations were likely subsequently founded from a reduced number of individuals.


| INTRODUC TI ON
Invasion biology has historically benefited from a partnership with population genetics to clarify evolutionary aspects of the establishment and spread of colonizing species (Allendorf & Lundquist, 2003;Baker & Stebbins, 1965;Barrett, 2015). Such interplay has motivated many genetics studies to identify the geographic origin of non-native populations, assess postcolonization gains or losses in genetic diversity, and illuminate population connectivity and genetic structure to provide strategies for control (Dlugosch & Parker, 2008;Roman & Darling, 2007;Sakai et al., 2001). Yet, there are many unresolved questions on how artificial and natural dispersal pathways differentially influence the population genetics of colonizing populations. The former relates to human-mediated releases or propagule pressure, intentional and unintentional, of non-native individuals into the receiving environment; the latter relates to spread of non-native individuals from established populations via gene flow. Populations founded via artificial dispersal may involve multiple introductions with expected increases in genetic diversity (Consuegra, Phillips, Gajardo, & de Leaniz, 2011;Kolbe et al., 2004;Simberloff, 2009), whereas natural populations may originate via founder effects and dispersal from established populations, exhibiting decreased genetic diversity (Kawamura et al., 2010;Kinziger, Nakamoto, Anderson, & Harvey, 2011;Rollins et al., 2013). However, little is known about how these two dispersal pathways of invasion differentially influence (i) genetic diversity and demography, namely the annual effective number of breeders (N b ), (ii) spatial patterns of genetic structure, and (iii) the degree of genetic admixture among non-native populations.
Pacific salmon species (genus Oncorhynchus) native to the Northern Hemisphere are important subjects in ecology and evolutionary biology because they exhibit diverse life histories (Quinn, 2005;Stearns & Hendry, 2004), and also because they have been successfully introduced around the world for commercial and recreational fisheries and aquaculture (Crawford & Muir, 2008).
Introductions of Pacific salmon into the Southern Hemisphere, especially New Zealand and South America, have provided unique research opportunities to investigate the relative roles of propagule pressure, preadaptations, phenotypic plasticity, and low ecosystem resistance in explaining invasions (Arismendi et al., 2014;Quinn, Kinnison, & Unwin, 2001). Several genetics studies in South America have ascertained the origin of donor (native) salmonid populations (Ciancio, Riva-Rossi, Pascual, Anderson, & Garza, 2015;Riva-Rossi, Lessa, & Pascual, 2004;Riva-Rossi et al., 2012). Other studies have shown significant gains in genetic diversity among non-native populations that may be important for invasion success (Correa & Moran, 2017;Di Prinzio, Rossi, Ciancio, Garza, & Casaux, 2015;Narum et al., 2017). Yet, answers to fundamental questions of how salmonids have established and spread throughout South America, a phenomenon that is invariably related to how and where they have been propagated (artificial vs. natural), are lacking. Anadromous salmonids that spawn in freshwater but feed in the ocean as adults are especially appropriate to study as they can rapidly colonize unoccupied habitats by dispersal. They can become quickly established, extending their distribution at their native (Hendry, Castric, Kinnison, & Quinn, 2004;Quinn, 2005) and non-native ranges (Quinn et al., 2001).
Chinook salmon (O. tshawytscha) is the largest-bodied Pacific salmon species, recognized for its importance for recreational, commercial, and subsistence fisheries in North America and Asia.
Chinook salmon were also repeatedly introduced to South America via several government-sponsored and private initiatives during most of the twentieth century, both in Chile and in Argentina, to develop recreational and commercial fisheries as well as net-pen  and natural and between natural populations, with strong differentiation between distant Pacific Ocean and Atlantic Ocean populations, which experienced strong genetic drift. Third, genetic mixture analyses revealed the presence of at least six donor geographic regions from North America, some of which likely hybridized as a result of multiple introductions. Relative propagule pressure or the proportion of Chinook salmon propagules introduced from various geographic regions according to government records significantly influenced genetic mixtures for two of three artificial populations. Our findings support a model of colonization in which high-diversity artificial populations established first; some of these populations exhibited significant admixture resulting from propagule pressure. Low-diversity natural populations were likely subsequently founded from a reduced number of individuals.

K E Y W O R D S
Argentina, Chile, genetic stock identification, individual assignment, invasion genetics, Pacific salmon aquaculture (Basulto, 2003;Pascual & Ciancio, 2007). Several studies indicate that successful introductions of Chinook salmon to South America, with adults returning in large numbers, occurred following sea-ranching experiments in the Lake (X Region) and Magallanes (XII Region) districts in Chile at the end of 1970s and beginning of 1980s (Correa & Gross, 2008;Niklitschek & Toledo, 2011;Riva-Rossi et al., 2012;Soto, Arismendi, Di Prinzio, & Jara, 2007). Many other populations have formed beyond initial and well-documented stocking sites, suggesting that Chinook salmon in South America comprise both artificial and natural populations (Table 1). Past and recent genetics and genomics studies have identified multiple donor populations of Chinook salmon that likely interbred and now coexist among Pacific Ocean and Atlantic Ocean basins (Correa & Moran, 2017;Narum et al., 2017;Riva-Rossi et al., 2012). Both individual assignment and genetic analysis of population mixtures (McKinney, Seeb, & Seeb, 2017) have greatly assisted tracking the geographic origin of donor Chinook salmon populations to various sites in South America Correa & Moran, 2017;Di Prinzio et al., 2015).
However, we lack a clear understanding on how genetic diversity, dispersal, and genetic admixture are linked to artificial and natural dispersal pathways of this species from its distributional range in South America. This is crucial to understand how colonization by non-native Chinook salmon has unfolded in less than 40 years (Correa & Gross, 2008;Riva-Rossi et al., 2012).
Here we analyzed nine Chinook salmon collections taken from artificial and natural populations (defined a priori) through their South American distribution, including Pacific Ocean and Atlantic Ocean basins (Figure 1), using a panel of 172 polymorphic SNPs.
These markers were developed from an ascertainment panel of wild and hatchery populations from the native range in North America (Warheit, Seeb, Templin, & Seeb, 2012) and have proven to be extremely informative among non-native populations. We used individual-and population-based inference, introduction records, and a baseline of genetic information from donor populations to address three goals in relation to the colonization history of Chinook salmon in South America. First, we quantified genetic diversity, contemporary dispersal, and N b among populations to test the prediction that artificial populations should harbor more genetic diversity and have larger estimates of N b than natural populations. We also tested whether contemporary dispersal was evident from artificial to natural populations, assuming the former were established first and the latter subsequently founded. Second, we evaluated genetic divergence and tested for genetic isolation by distance. We predicted that genetic differentiation may be weaker between adjacent artificial and natural populations assuming ongoing gene flow, but stronger between distant pairs, especially if genetic drift strongly influenced natural populations. Third, we inferred which donor (native) populations contributed to establishment of non-native Chinook salmon using analyses of genetic mixtures. We predicted whether genetic mixtures were consistent with relative propagule pressure or the proportion of Chinook salmon propagules introduced from various geographic regions according to historical government records.

| Sampling
We covered the entire distributional range of Chinook salmon among Pacific Ocean and Atlantic Ocean basins in South America by supplementing archived samples of seven populations (Riva-Rossi et al., 2012)

| SNP genotyping
Genomic DNA was isolated using a Macherey-Nagel NucleoSpin ® Tissue kit (Düren, Germany) following the protocols from the manufacturer. Some isolates from Chinook salmon carcasses contained low concentrations of DNA (<10 ng/μl), and these generally yielded low-quality genotypes that were excluded from analyses. DNA was screened for a suite of 191 SNPs (Table S1) chosen from a larger database (288 SNPs) developed to coordinate genomic resources available for improving Chinook salmon fisheries management (Warheit et al., 2012). Exploratory analyses showed that different suites of SNPs had similar information content and performed equally well for fisheries applications (Warheit et al., 2012). Genotyping was performed on Fluidigm ® 96.96 dynamic arrays under PCR conditions and concentrations recommended by Seeb et al. (2007), following a preamplification step by Smith et al. (2011).

| SNP selection
We tested whether our data fit Hardy-Weinberg equilibrium (HWE) proportions or showed evidence of linkage disequilibrium (LD) using GENEPOP v4.2 (Raymond & Rousset, 1995b;Rousset, 2008) through 10,000 dememorization steps, 100 batches, and 5000 iterations per batch, using the complete enumeration method. SNPs that consistently departed from HWE proportions or were found in LD in more than half of locations were excluded from subsequent analyses.

| Genetic diversity, N b , and gene flow
We estimated observed (H O ) and expected heterozygosity (H E ) for each collection using GENALEX (Peakall & Smouse, 2012)  The effective number of breeders per population (N b ) was calculated in NeEstimator using the linkage disequilibrium (LD) approach (Do et al., 2014;Waples & Do, 2010). The method has the important advantage of requiring a single sample to providing unbiased estimates of contemporary N b per brood year in age-structured populations, including semelparous salmonids with variable age at maturity. We ran NeEstimator on these settings: (i) threshold values for minor allele frequencies dependent on sample sizes n ≥ 25 F I G U R E 1 Distribution of non-native Chinook salmon sampled locations (circles) in South America. Light gray, Pacific Ocean basins; dark gray, Atlantic Ocean basins or n < 25, (ii) random mating system, and (iii) jackknifing to calculate 95% CIs (Waples & Do, 2010). We tested whether LD N b differed between artificial and natural populations using a nonparametric Mann-Whitney test in R.
First-generation immigrants were identified using GENECLASS 2.0 (Piry et al., 2004) to gauge contemporary gene flow between artificial and natural populations. We used Bayesian assignment of genotype likelihoods following Rannala and Mountain (1997).
Genotypes were then ranked according to Paetkau, Slade, Burden, and Estoup (2004) in order to relate the likelihood of drawing them from the populations in which they were sampled with the maximum likelihood of such genotypes considering any of the study populations. The method assumes that all sources of migrants have been sampled. This may not be the case and "ghost" unsampled populations may influence these analyses, suggesting estimates of gene flow and identification of immigrants need to be interpreted with caution. We used 10,000 simulations and set type I error α = 0.01 to minimize erroneous identification of immigrants.

| Genetic structure and isolation by distance
First, we used an individual-based discriminant analysis of principal components (DAPC) (Jombart 2008) on multilocus genotypes to analyze genetic structure. Clustering of individuals was performed by maximizing their genetic proximity using the k-mean algorithm. We varied k or the number of clusters between k = 2 and k = 9 and tested the significance of each k using the first 150 principal components and Bayesian information criterion.
Second, we estimated pairwise genetic divergence between populations using Weir and Cockerham (1984) θ estimator. Exact tests of population differentiation for the null hypothesis θ = 0 (Raymond & Rousset, 1995a) were applied using GENEPOP and 10,000 dememorization steps, 100 batches, and 1000 iterations per batch. We used Fisher's method in GENEPOP to combine exact tests per locus over multiple loci. We also tested whether pairwise genetic and geographic distances were significantly correlated using a sim-

| Simulations
First, we tested for the accuracy afforded by baseline SNP genotypes among geographic regions to correctly assign simulated mixtures using ONCOR (Kalinowski, Manlove, & Taper, 2008)

| Membership probabilities and mixture proportions to donor geographic regions
We performed mixture analysis in ONCOR to estimate mixture proportions of geographic regions using conditional maximum likelihood (Millar, 1987). We additionally estimated individual membership probabilities to each of 14 regions using Bayesian assignment (Rannala & Mountain, 1997) in ONCOR. Individual membership probabilities were represented as stacked bar plots in R, with colors depicting regions.

| Relative propagule pressure
We compared results from genetic mixtures to relative propagule pressure defined as the proportion of donor regions identified on government records. We surveyed information from three well-

| SNP selection
We successfully genotyped 342 Chinook salmon from nine locations (

| Genetic diversity, N b , and gene flow
Values for H O and H E varied similarly across populations (Table 2); We identified 14 first-generation immigrants at the p < .01 threshold (Table 3). Gene flow between adjacent populations (10 of 14 immigrants) occurred from artificial to natural populations as predicted (COB to VAR) as well as in the opposite direction (VAR to PRA). We also observed dispersal among adjacent natural populations (SER and VAR; SAC and CAT). We observed additional long-distance, bidirectional dispersal with four immigrants between COB in the Pacific Ocean and CAT in the Atlantic Ocean, two populations separated by nearly 4000 km of coastal distance. We found no evidence for connectivity between northern (ALP and PET) and southern populations (all the rest), other than one possible immigrant from PIC (a hatchery population) to PET.  Genetic distance values fluctuated between θ = 0.011 (COB vs.

| Genetic structure and isolation by distance
VAR) and θ = 0.231 (PIC vs. CAT) and were all significantly higher than 0, suggesting weak to strong spatial genetic structure (Table   S2). A simple Mantel test revealed that θ and geographic distances were significantly correlated (r = .613, p < .001; Figure 4a). However, it became evident that the significance of this relationship was chiefly driven by pairwise comparisons between artificial and natural populations, and between natural populations. Comparisons between artificial populations showed no significant relationship between θ and geographic distances (r = .19, p > .10; Figure 4b).

| Genetic mixtures and relative
propagule pressure

| Baseline of donor populations and simulations
There was significant overlap between donor and non-native suites of SNPs, although they were not identical. We found 127 SNPs in the non-native populations that matched the donor population database (Table S1); only these were subsequently used for these analyses. First, simulated mixtures created from the baseline of native populations were correctly assigned to their geographic region in 99% or 100% of cases, strongly suggesting estimation of mixture proportions and individual assignment was highly accurate using 127 SNPs (Table S3). Second, F1, F2, and backcrosses simulated in HYBRIDLAB were reliably assigned to two selected parental geographic regions (Oregon-California Coast and Lower Columbia River-Willamette) with only 1 of 200 backcrossed individuals misassigned to a third, and closely related, parental geographic region (Columbia River-Deschutes: Figure S1).  Figure S2).

| Relative propagule pressure
The highest propagule pressure invariably originated via Seattle in Washington, USA, for basins PET and PRA, and via Vancouver in British Columbia, Canada, for COB, with both locations of origin of propagules assigned to Puget Sound-South British Columbia (Table 5). Introduction records further showed other locations in North America, including Lower Columbia River and Oregon hatcheries, and other countries such as New Zealand, which was only found among introductions to PET (Table 5). We found that genetic mixture proportions were consistent with relative propagule pressure proportions estimated for PET (Pearson χ 2 = 0.223, p = .974) F I G U R E 4 Scatterplot between linearized genetic and geographic distances among non-native Chinook salmon populations from South America and best-fit regression lines from two datasets: (a) pairwise comparisons among all nine populations and (b) pairwise comparisons between artificial populations (artificial-artificial; best-fit, dotted line), between artificial and natural populations (artificial-natural; best-fit, n-dash line), and between natural populations (natural-natural; best-fit, m-dash line)

| D ISCUSS I ON
We used multilocus genotypes (from both non-native and native and donor populations), historical records of introduction, and various statistical approaches to evaluate the roles of artificial and natural dispersal pathways of Chinook salmon (Oncorhynchus tshawytscha) in South America using a population genetics approach. Although Chinook salmon show strong philopatry, they also colonize new habitats via dispersal. Our genetic survey of nine non-native populations covered nearly 5500 km and provided resolution across the distributional range to evaluate how colonization of this species has unfolded during almost four decades. Below we discuss how our findings informed and supported predictions regarding genetic diversity, the evolution of genetic structure, and population admixture. were higher for non-native rainbow trout populating a Patagonian lake with intensive trout aquaculture than for a lake where trout aquaculture has been prohibited by law. This highlights a possible role of continuous trout escapes and artificial dispersal on enlarging N b . Praebel, Gjelland, Salonen, and Amundsen (2013) (Waples, 2005). Second, most adult samples included one or two consecutive brood years, with exception of CAT that included multiple (pooled) brood years; also, smolt samples from SAC were pooled across 5 years. Estimates for these populations approach the effective population size (N e ) rather than N b as samples span one entire generation of Chinook salmon (Waples, 2005;Waples & Do, 2010).

| Genetic diversity, N b , and gene flow
Patterns of contemporary gene flow indicated that dispersal among adjacent populations occurred in all possible directions.
We observed gene flow from artificial to natural populations (COB to VAR in the Pacific Ocean), suggesting that artificial populations were likely established first and natural populations were subsequently founded (VAR to PRA in the Pacific Ocean).
We also observed gene flow between natural populations (VAR to SER in the Pacific Ocean; SAC to CAT in the Atlantic Ocean).
Further, we found evidence for long-distance dispersal between Pacific Ocean and Atlantic Ocean basins (COB to CAT and vice versa). Both strategies may be common among successful invasions (Wilson, Dormontt, Prentis, Lowe, & Richardson, 2009).
Long-distance dispersal in particular appears crucial for establishment and spread of invasive populations in both theoretical (Ibrahim, Nichols, & Hewitt, 1996;Shigesada, Kawasaki, & Takeda, 1995)  Both adjacent and long-distance dispersal strategies appear to be important for establishment and spread of Chinook salmon, similar to observations in other invasive fishes (Bronnenhuber et al., 2011). Unwin and Quinn (1993) reported straying rates 4-20% among non-native Chinook salmon in New Zealand, with most occurring proximate to the tagging site. However, habitat variation causes salmon to stray selectively, so proximity is not the only factor influencing dispersal (Pascual & Quinn, 1994;Westley, Dittman, Ward, & Quinn, 2015). Additionally, the influence of oceanographic currents flowing southward along the southeastern Pacific Ocean

| Genetic structure and isolation by distance
DAPC identified three clusters consistent with independent stocking events among artificial populations in Pacific Ocean basins (e.g., ALP, PET, PRA), even though some amount of gene flow has occurred between artificial (COB, PRA) and natural (VAR, SER) populations, especially south of 45°S. Also, PRA was stocked with Chinook salmon that likely founded PET (Correa & Gross, 2008;Riva-Rossi et al., 2012), which explains why PRA and PET genotypes partially overlap in DAPC analyses despite the fact that nearly 3,000 km separate the two populations. A fourth cluster comprised Chinook salmon from SAC and CAT, two natural Atlantic Ocean populations that showed strong differentiation (θ = 0.069-0.231; Table S2) from all Pacific Ocean basins. Results from two mitochondrial DNA studies (Becker et al., 2007;Riva-Rossi et al., 2012) and another using biparental SNPs  failed to support successful stocking of Chinook salmon from California, United States, to Atlantic Ocean basins at the beginning of the twentieth century. On the contrary, those studies supported the origin of propagules from Pacific Ocean basins during more recent periods. Sea-ranching operations that released juvenile Chinook salmon at PRA from 1982 to 1988 were likely the source of founders for Santa Cruz River (SAC), wherein spawning grounds are found at CAT.
The first records of Chinook salmon by anglers at CAT dated back to 1979-1984, supporting the Pacific Ocean origin hypothesis (Ciancio, Pascual, Lancelotti, Rossi, & Botto, 2005). A strong founder effect and subsequent genetic drift are possible explanations as to why SAC and CAT strongly diverged from Pacific Ocean basin sources. Yet, selection cannot be discounted as another explanation for divergence, especially among colonizing populations (Hanfling, 2007;Lee, 2002). Narum et al. (2017) examined adaptive genomic variation by comparing native and non-native Chinook salmon sampled from Patagonia and found evidence for 118 outlier SNPs that deviated from neutral expectations (1%; 11,579 SNPs in total). Some of these outliers were linked to immune function, transposons, regulation of transcription, and histone acetylation (Narum et al., 2017). This is consistent with the concept of "favored founders"; they are not only a small subset of the source population, but a nonrandom, preadapted subset because they had to survive and reproduce to establish the population (Quinn et al., 2001). We therefore encourage further investigations on how adaptive divergence explains successful invasions, currently an active area of research on the use of thousands of SNPs to investigate genome signatures of selection (Puzey & Vallejo-Marín, 2014;Vandepitte et al., 2014;White, Perkins, Heckel, & Searle, 2013).
Isolation by distance among invasive populations may evolve from genetic differentiation following geographic expansion from a single source (Herborg, Weetman, Van Oosterhout, & Hanfling, 2007;Kawamura et al., 2010;Kinnison, Bentzen, Unwin, & Quinn, 2002) or secondary contact between multiple introductions (Bifolchi, Picard, Lemaire, Cormier, & Pagano, 2010). None of these explanations seem to apply to Chinook salmon in South America. We hypothesize that this pattern has emerged from weak differentiation between adjacent artificial and natural populations as well as between natural populations, combined with strong differentiation between geographically distant populations influenced by genetic drift, namely Atlantic Ocean populations.
This stems from the fact that artificial populations made no contribution to differentiation. performance (Narum et al., 2008).

| Genetic mixtures and relative
Both individual assignment and genetic mixtures supported a broad geographic origin of donor populations spanning six geographic regions, consistent with introduction records and the history of propagules imported to South America (see Appendix 1 for additional details) as well as a parallel study employing microsatellite markers on some of the same locations (Correa & Moran, 2017).
Some uncertainty was associated with estimation of uncommon donor geographic regions in mixtures, reflected in wide 95% CIs, some of which contained zero as lower bound. A diverse origin of propagules contrasts strongly with the Chinook salmon gene pool in New Zealand, the most closely studied comparable case, where the species seems to have originated from a single source from California (USA), and dispersed largely by natural reproduction (McDowall, 1994;Quinn et al., 1996). The New Zealand scenario resembles natural populations in our study as they were composed of a What is the explanation behind the predominance of Lower Columbia River-Willamette geographic region? Successful invasions are often a combination of stochastic and directional forces (Keller & Taylor, 2008). We speculate that genetic drift, selection on existing preadapted genes, or both, may be potential explanations. Selection on preadapted genes or the "preadaptation hypothesis" implies that specific populations will be successful only on specific environments (Chown et al., 2015;Sax & Brown, 2000), and Chinook salmon lineages may be no exception. Narum et al. (2017) compared native and non-native Chinook salmon environments using high-resolution global climate layers. They concluded that non-native environments had higher precipitation and lower temperatures; they also found variation among Patagonian sites located at various latitudes (Narum et al., 2017). Thus, whether temperature, precipitation, flow regime, or migration distance from sea among South America basins have provided opportunities for selection of specific Chinook salmon lineages deserves further scrutiny by contrasting genetic and phenotypic data from both native and non-native populations.
We found evidence for hybridization between non-native Chinook salmon lineages as several individuals from artificial populations were assigned to two or more geographic regions. This was also reported by Correa and Moran (2017). It is unclear whether hybridization between lineages occurred in captivity prior to stocking, progressively in the wild following multiple stocking events or aquaculture escapes, or all the above. No significant departures from HWE proportions and linkage equilibrium within collections suggest that perhaps enough generations have passed as to dissipate Wahlund effects. However, whether admixture enhances invasiveness among non-native populations is still unresolved (Hahn & Rieseberg, 2017;Rius & Darling, 2014;Wolfe, Blair, & Penna, 2007).

| Relative propagule pressure
Propagule pressure may help mitigate demographic, environmental, and genetic stochasticity among non-native populations, and can thus be a key factor explaining why species become invasive (Simberloff, 2009). Although propagule pressure was shown to increase genetic diversity among artificial populations in this study and elsewhere (Consuegra et al., 2011;Roman & Darling, 2007), little is known about how relative propagule pressure may influence genetic mixtures. We found consistency between genetic mixtures and relative propagule pressure inferred from introduction records of Chinook salmon for two of three basins analyzed, Petrohué River (PET) and Prat River (PRA). These findings suggest that the more propagules were introduced from a specific geographic region, the larger its genetic contribution, in line with demographic effects of propagule pressure during the establishment phase of an invasion (Szűcs, Melbourne, Tuff, & Hufbauer, 2014 We additionally found that measures of relative propagule pressure and mixture analyses for Chinook salmon populating Cobarde River (COB) were inconsistent, suggesting that the demographic effects of propagule pressure cannot be generalized, and that history and chance or deterministic factors may affect invasion success (Keller & Taylor, 2008

| CON CLUS ION
Artificial and natural dispersal pathways left unique signals on genetic metrics, the evolution of genetic structure, and degree of admixture among non-native Chinook salmon populations in South America.
Artificial populations had higher genetic diversity and larger estimates of N b than natural populations. Gene flow seemed more common between adjacent artificial and natural, or adjacent natural, than geographically distant populations. Findings were consistent with a process of colonization in which high-diversity artificial populations likely established first followed by founding of low-diversity natural populations. Genetic mixtures helped identifying donor geographic regions and the influence of relative propagule pressure. Overall, the study of non-native Chinook salmon mixtures in combination with well-documented introductions and their origin represents a unique approach to study differential invasion success among genetically distinct donor populations.

CO N FLI C T O F I NTE R E S T
None declared. We discussed in further detail the origin of Chinook salmon introductions from introduction records and present-day inferences of population mixtures. We matched evidence from these two sources among stocked (artificial) populations from our study and present additional details on how we assigned the hatchery of origin of Chinook salmon propagules (often eggs that were hatched and reared to juveniles before stocking) to geographic regions.  (Zubillaga, 2013). According to Zubillaga (2013), hatchery company owners released their broodstock to ALP in 1995 after filing for bankruptcy. Since 2000, large numbers of Chinook salmon adults became common on spawning grounds in several of ALP's tributaries. Strong genetic divergence between ALP and the rest of Chinook salmon populations located further south assessed from various methods supports the hypothesis of an independent, artificial introduction in the Toltén River basin. Evidence from genetic mixtures is nonetheless in conflict with anecdotical data regarding the hatchery of origin of Chinook salmon released to ALP. Most were assigned to

A LLI PE N ( A LP) R I V E R
Oregon-California Coast (46%), followed by Lower Columbia River-Willamette (36%), and none to California Central Valley (Table 4). This is consistent with large numbers of eggs imported to Chile's Lake District from Oregon during 1990s (Table 5). All Oregon State hatcheries can be located within geographic region Oregon-California Coast (Warheit et al., 2012).

Oregon-California Coast via Oregon, and California Central Valley via
New Zealand (Correa & Gross, 2008;Donaldson & Joyner, 1983;Niklitschek & Toledo, 2011;Quinn et al., 1996;Riva-Rossi et al., 2012). Genetic mixture analysis suggests contributions of both early sea-ranching and late aquaculture broodstock from various geographic sources. They contributed with founding individuals to PET and nearby basins, such as Futaleufú River (Di Prinzio et al., 2015).

E S TE RO PI CH I CO LO (PI C)
It harbors one of the few Chinook salmon hatcheries in Chile. The origin of the broodstock was Washington State, most likely from University of Washington (Riva-Rossi et al., 2012). Mixture analysis shows a large contribution from Puget Sound-South British Columbia (86%) which is consistent with this prediction. However, contributions from Columbia River-Deschutes (11%) and California Central Valley (3%) were also found, suggesting admixture with other geographic regions. Astorga, Valenzuela, Arismendi, and Iriarte (2008) found significant genetic divergence between PIC and PET using three microsatellite DNA markers, and argued that sources other than PIC were likely founders of Chinook salmon at PET. Individualbased analyses in this study showed on the contrary that these two populations clustered within the same group and genetic differentiation was weak, albeit significant (θ = 0.03; Table S2). These findings imply that their origin is similar and can be tracked back to University of Washington broodstock, though escapes from PIC that may have contributed to PET cannot be ruled out.

CO BA R D E (CO B) R I V ER
Historical records indicated that Chinook salmon broodstock used in net-pen aquaculture near Magdalena Island and Aysén River basin were imported from Vancouver, British Columbia (Niklitschek & Toledo, 2011;Riva-Rossi et al., 2012).

PR AT (PR A ) R I V E R
The second basin in South America to be successfully stocked