Genomics-informed models reveal extensive stretches of coastline under threat by an ecologically dominant invasive species

Significance

Species with narrow distributions provide unique opportunities for understanding the mechanisms that limit their spread. We studied a marine invader that exhibits ecological dominance within its range and has the capacity to fundamentally alter the coastal habitat when introduced to new locations. We found evidence of the species’ potential to establish itself far beyond its present introduced range from both genomic data and species distribution modeling. Therefore, minor oceanographic changes (due to, for example, contemporary climate change) or alteration to human-mediated dispersal may trigger a large-scale invasion along vast stretches of coastlines. Our work provides a holistic framework to assess potential changes in the distribution of invasive species.

Abstract

Explaining why some species are widespread, while others are not, is fundamental to biogeography, ecology, and evolutionary biology. A unique way to study evolutionary and ecological mechanisms that either limit species’ spread or facilitate range expansions is to conduct research on species that have restricted distributions. Nonindigenous species, particularly those that are highly invasive but have not yet spread beyond the introduced site, represent ideal systems to study range size changes. Here, we used species distribution modeling and genomic data to study the restricted range of a highly invasive Australian marine species, the ascidian Pyura praeputialis. This species is an aggressive space occupier in its introduced range (Chile), where it has fundamentally altered the coastal community. We found high genomic diversity in Chile, indicating high adaptive potential. In addition, genomic data clearly showed that a single region from Australia was the only donor of genotypes to the introduced range. We identified over 3,500 km of suitable habitat adjacent to its current introduced range that has so far not been occupied, and importantly species distribution models were only accurate when genomic data were considered. Our results suggest that a slight change in currents, or a change in shipping routes, may lead to an expansion of the species’ introduced range that will encompass a vast portion of the South American coast. Our study shows how the use of population genomics and species distribution modeling in combination can unravel mechanisms shaping range sizes and forecast future range shifts of invasive species.

Fundamental to biogeography, ecology, and evolutionary biology is understanding why some species are widespread, whereas others are not (1). Studies focusing on species ranges often use predictive modeling to infer the potential spatial spread of species (2), with a growing number of studies comparing fundamental niches (i.e., the entire set of conditions that a species can tolerate) and realized niches (the actual set of conditions under which a species is found) (3–5). In recent times, our understanding of range sizes has advanced considerably thanks to integrative studies (6, 7) that have provided key insights into how local adaptation (8), physiological tolerance (9), and propagule dispersal (10) shape population persistence and spread potential. Despite all this progress, our understanding of how ecological and evolutionary mechanisms shape range sizes and niche occupancy remains limited (11–13).

Nonindigenous species (NIS) offer unique opportunities to study range sizes, especially when they establish and spread into new geographic areas. Most studies to date have focused on highly invasive species that have already had significant ecological and economic impacts (14, 15), and that have spread over large areas in their introduced range (16). In turn, relatively little research has been afforded to invasive species with restricted introduced ranges, or naturalized species (i.e., species that establish self-sustaining populations beyond their native range, but that have not yet expanded from their point of introduction) (17–20). These species represent unique systems to study mechanisms responsible for limiting NIS spread, such as biotic resistance (21, 22) and genetic bottlenecks (23), and/or facilitating range expansions (24).

A rarely used approach to study NIS is the use of species distribution models (SDM) and population genomics (25) in combination. SDM have become widely used to both identify regions of suitable habitat across landscapes (26) and to predict areas at risk of future range shifts (27, 28). In addition, recently developed analytical tools in population genomics offer powerful ways of studying demographic history, fine-scale population structure, adaptive divergence (29), and eco-evolutionary processes associated with NIS’ range shifts (30, 31). Therefore, the use of both genomic data and SDM has the potential of improving our ability to characterize the mechanisms that shape range sizes. This includes allowing high-resolution spatial delineation of population structure (32), identification of landscape elements that drive ecological and evolutionary patterns (33), and substantially improving our ability to predict future range shifts (25, 34).

Here, we used population genomics and SDM to investigate mechanisms that shape the distribution of the ascidian Pyura praeputialis (Heller, 1887), a sessile benthic marine invertebrate that has proven to be an aggressive invader, but that has a geographically constrained introduced range. In addition, we evaluated the potential for the spread of this species to adjacent areas. We specifically aimed to 1) understand the nature of recent range size changes through the reconstruction of the species’ invasion history, 2) determine the occupancy levels of suitable habitat within and beyond the current native and introduced ranges, and 3) evaluate whether our data can explain the current constrained distribution of the study species and predict future spread. We expected that only a subset of the genomic diversity from the native range would be present in the introduced range, potentially indicating the presence of a genetic bottleneck. Additionally, we predicted that our genomics-informed SDM would reveal suitable habitat along adjacent coastlines of the current introduced range, but that lack of adaptive capacity, as revealed by low genomic diversity in introduced populations, would limit its spread.

Results

Processing of Raw GBS Data.

We retained a total of 1,205 putatively unlinked single-nucleotide polymorphisms (SNPs), with the final dataset comprising 164 individuals from 13 sampling sites (SI Appendix and Table S1). A total of 49 candidate adaptive loci were identified by bayenv2 and 30 by redundancy analysis (RDA) (SI Appendix, Fig. S1), with 14 loci shared between the two methods (for a full description of the genotype–environment association [GEA] analyses, see SI Appendix). We therefore treated the 65 loci that were retained by either method as candidate adaptive loci (hereby called the “candidate dataset”) and generated a “neutral dataset” using the remaining 1,140 putatively neutral SNPs.

Population Structure and Reconstruction of Invasion Routes.

ADMIXTURE, discriminant analysis of principal components (DAPC), and F_ST (SI Appendix, Figs. S2–S4) analyses based on both the neutral and candidate datasets all supported the existence of two highly differentiated groups of populations within Australia. The individuals from Antofagasta Bay in Chile were recovered in the same cluster as those from the eastern Australian sites (Fig. 1). When southeastern Australian populations were removed from the DAPC, the populations from Antofagasta Bay and eastern Australia still clustered together (SI Appendix, Fig. S3 C and D). In contrast, there was no fine-scale genomic structure within the introduced range (SI Appendix, Fig. S3 E and F). This was also the case when only candidate loci associated with sea surface temperature (as inferred from the RDA) were used (SI Appendix, Fig. S5). Similarly, approximate Bayesian computation (ABC) analyses showed that Antofagasta Bay was most likely founded by individuals from eastern Australia (probability P = 1.000, 95% CI = 1.000, 1.000; SI Appendix, Table S2 and Fig. S6), rather than from southeastern Australia or from an admixture of the two Australian lineages (SI Appendix, Table S2). In addition, these simulations suggested that the effective population size introduced to Chile from eastern Australia was of the order of thousands (SI Appendix, Fig. S7).

Fig. 1.

Sampling sites of Pyura praeputialis along the coast of (A) Australia and (B) within Antofagasta Bay on the Chilean coastline. The bar plots in A and B represent the results of the ADMIXTURE clustering analysis inferred with neutral loci at K = 2 (see full details in SI Appendix). (C) The most likely scenario, as revealed using approximate Bayesian computation (see full details in SI Appendix), depicts the most likely evolutionay scenario and invasion route that P. praeputialis followed from Australia to Antofagasta Bay.

Species Distribution Modeling.

When models were produced using the appropriate genetic lineage of P. praeputialis (i.e., genomics-informed; see Fig. 2), coastlines adjacent to and far beyond the introduced range of Antofagasta Bay were found to be suitable habitat for this species. These models showed a dominating effect of the variable “distance to shore,” with “maximum sea surface temperature” being the second most important variable (SI Appendix, Table S3). Upon removing the distance to shore variable, we observed models where the variables “maximum current velocity” and “maximum sea surface temperature” were the most dominant (SI Appendix, Table S4 and Fig. S8). Finally, SDM produced excluding the introduced range and built using both native lineages (i.e., not genomics-informed; see Fig. 3) did not recover vast stretches of coastline as suitable for P. praeputialis.

Fig. 2.

Habitat suitability for Pyura praeputialis across (A) its native range, (B) the coastline adjacent to Antofagasta Bay (Inset; Antofagasta Bay indicated with an arrow), and the western coastline of South America. The scale bar represents Maxent’s logistic output, with “yellow” indicating high habitat suitability. Note that the Maxent’s logistic output only considers the genomics-informed relevant sites (i.e., sites from Chile and the eastern coastline of Australia; see details in main text and SI Appendix). This model includes distance to shore as a variable, explaining the observed narrow regions of suitable habitat. For a full list of variables used in model creation, see SI Appendix, Table S3.

Fig. 3.

Maps illustrating habitat suitability for Pyura praeputialis when the introduced range is not considered in model construction. The figures show the output of Maxent modeling using occurrence data from: both native lineages (A–C), only the southeastern Australian lineage (D–F), and only the eastern Australian lineage (G–I). Maps depict native range (A, D, and G), adjacent coastlines of Antofagasta Bay (bay represented by arrow) (B, E, and H), and extensive coastlines along the western coast of South America continent (C, F, and I). The scale bar represents Maxent’s logistic output on habitat suitability (see details in SI Appendix).

Discussion

Our study shows how using SDM and population genomics in combination can refine both our understanding of mechanisms responsible for range size changes and our predictions of spread potential of a regionally constrained NIS. We first found genomic evidence of considerable adaptive potential in the highly restricted introduced range, suggesting great potential for spread. In addition, our analyses revealed a large and genetically diverse founder population, which is in line with historic high levels of artificial transport between the native and introduced ranges (see SI Appendix). Although only one of two lineages contributed to the successful introduction of this species, no footprint of a genetic bottleneck could be found. With this information, our SDM showed that over 3,500 km of coastline along the eastern Pacific is potentially at risk of invasion. The aggressive invasive behavior shown by this bioengineering NIS within its introduced range suggests that if this species expands its range, it could potentially threaten marine ecosystems along the coastline of northern Chile, the entire coastline of Peru, and much of the coast of Ecuador.

Our genomic dataset confirmed the presence of two native lineages of P. praeputialis and no evidence of finer-scale cryptic population structure, as previously suggested using a limited number of genetic markers (35, 36). We found that southeastern Australian populations did not contribute to the invasion of P. praeputialis in Chile, confirming a single-lineage introduction to Chile (i.e., the lineage found on the east coast of Australia; Fig. 1C) (35, 37). The southeastern Australian lineage of P. praeputialis also inhabits a region with busy ports (e.g., Melbourne), and one would expect that this region would also have been linked to South America through historical shipping. Species that occur in multiple biogeographic zones are often subdivided into distinct evolutionary lineages that are adapted to regional conditions (38), and movement to other biogeographic zones can disrupt growth, reproduction, development, and survival of the migrants (39, 40). Such effects may have limited the ability of the southeastern Australian lineage to colonize South America, an interpretation supported by the limited suitability of the Chilean coastline for this lineage inferred by our SDM (Fig. 3E). The first introduction of P. praeputialis to Chile would most likely have occurred during the late 19th century when maritime traffic was not only considerable between Australia and Chile, but when fouling organisms were rarely removed from the wooden hulls of ships (41).

High standing genomic diversity is expected to be the primary cause of adaptive potential in NIS (42) in both terrestrial (43, 44) and aquatic (45) environments. We found similar levels of genomic diversity across the range of P. praeputialis (SI Appendix, Table S5), and thus such observed levels of genomic diversity within the introduced range were not consistent with our initial hypothesis of a genetic bottleneck. Instead, these showed evidence of multiple introductions (46). Our results therefore suggest that, if the species was to spread beyond Antofagasta Bay, it may have sufficient genomic variation to adapt to the variety of conditions found along the South American coastline.

Research on local retention of marine organisms has shown that both active and passive mechanisms can promote such retention, including nutrient composition (47), odor cues (48), and hydrodynamic processes (49). Previous research has shown that idiosyncratic characteristics of both Antofagasta Bay and P. praeputialis facilitate retention of this ascidian within the bay. For example, an “upwelling shadow” (sensu ref. 50) is present within Antofagasta Bay, leading to stratification due to a shallow thermocline, cyclonic circulation, and high retention of water, with the existence of a persistent warm-water patch (51–53). This patch is generally found within the bay immediately downwind of an upwelling center, with temperatures on average 2–3 °C warmer inside the bay than outside (52, 53). Such phenomena are not limited to Antofagasta Bay, with similar features exhibited in the California current system (e.g., Monterey Bay and the Gulf of the Farallones) where areas of larval retention lead to distinct zooplankton assemblages over distances of just a few kilometers (52). In the particular case of Antofagasta Bay, the geometry of the coastline further aids in trapping surface water within the northern portion of the bay, creating an “upwelling trap” (52) where the surface waters are retained for several days. This upwelling trap has been proposed as a key retention mechanism for the planktonic larvae of P. praeputialis, which remain pelagic for less than 3 h (54). In contrast, other gregarious intertidal organisms found within Antofagasta Bay, such as the mussel Perumytilus purpuratus, have longer pelagic durations and extensive distributions along the west coast of South America (55). In addition to the short pelagic life-history stage of P. praeputialis, gametes and larvae can be retained by bio-foam produced by P. praeputialis adults (56), which further limits the dispersal of this species along the coastline outside of Antofagasta Bay (52, 56).

Improved predictive power of SDM requires input from both native and introduced ranges (57), but this is not feasible for NIS that are either cryptic or occupy limited areas in the introduced range and thus remain unsampled. Genetic identification of source populations has previously been used to inform climatic niche shifts (25), and our results showed the need for knowledge on cryptic genetic diversity to accurately predict potential range expansion. The results of SDM using occurrence data from both the native and introduced ranges of the genomics-informed source lineage indicate that the species has the potential to inhabit a much more extensive area along the southeastern Pacific coastline than is currently observed (Fig. 2). However, using only occurrence data from the eastern Australian lineage (i.e., excluding the introduced range), Antofagasta Bay was considered unsuitable for P. praeputialis (Fig. 3H). Furthermore, when the information on the exact source of the invasion was not considered, but the species’ complete native range or only its southeastern Australian range (i.e., nonsource native lineage) were used without occurrence data from the introduced range, the Chilean coast was not identified as suitable habitat (Fig. 3 B and E). Although our SDM incorporated a wide variety of environmental variables, future studies could consider other important variables such as microhabitat characteristics and biotic interactions (57).

Despite the lack of evidence of naturalization or spread of P. praeputialis to sites outside of Antofagasta Bay, work conducted by Castilla et al. (58) has shown that juvenile P. praeputialis can be transplanted outside of Antofagasta, and continue to survive and grow. Additionally, these authors concluded that there is no obvious biotic rationale, such as predator intensification or competitive exclusion, for the lack of expansion from Antofagasta Bay (58). This is in contrast to other NIS, which show limited introduced ranges due to predation (21), genetic bottlenecks (23), or population divergence (45). Successful transplantation of species outside of their ranges, overcoming niche constraints, is not rare (59) and shows that dispersal is an important explanatory variable of range limits. Therefore, it may be possible that Allee effects (60), the positive relationship between mean fitness and population density [e.g., mate limitation in broadcast spawners (61)], limit the ability of P. praeputialis to colonize naturally outside of Antofagasta Bay. Our genomic data suggested limited structuring within Antofagasta Bay, with sites closest to the mouth of the bay (C2 and C6, SI Appendix, Table S1) genetically separated in the DAPC results using loci associated with sea surface temperature (SI Appendix, Fig. S5). The warmest site within Antofagasta (C1) is also the region with the largest individuals and highest biomass, while the site furthest to the south of the bay (C6) has the lowest population density (53, 62). This in combination with larval retention mechanisms suggests that, should the abiotic conditions present in Antofagasta Bay (upwelling trap, subsurface water bringing in water rich in chlorophyll, bay retention mechanisms, etc.) change, P. praeputialis has great potential for range expansion and to subsequently alter biodiversity along and extensive stretch of the eastern Pacific coastline.

Studying failed introductions is inherently difficult as there is often no footprint left of the introduction event (63). It would be unreasonable to assume that P. praeputialis exclusively attached itself to ships that traveled between Australia and Antofagasta Bay (for more information, see text in SI Appendix). Despite this, we found no evidence behind an abiotic rationale limiting the colonization of additional regions along the eastern Pacific coastline. This suggests that previously failed introductions may have been due to variables not included in our analyses, changes in environmental conditions since the 19th century, or through limited propagule pressure or opportunity (64). The jump between introduction to a limited geographic area and widespread invasiveness has previously been linked to increased residence time (65). Indeed, time since invasion has been reported as the best predictor of range sizes in marine invertebrates (66), presumably enabling species to overcome the characteristic lag period of biological invasions (67). As there has been no expansion of P. praeputialis along the South American coast in more than 100 y, we nonetheless urge caution against complacency when monitoring NIS with restricted distributions. Recreational boating with poorly maintained hulls or sea-chests could spread adult individuals around, providing an opportunity for P. praeputialis to escape the unique oceanographic conditions of Antofagasta Bay.

In summary, we showed how population genomics and SDM can provide key insights into mechanisms shaping range sizes. We found that the invasive P. praeputialis has great potential for spread beyond its point of introduction and thus threatens coastal biodiversity along a large stretch of South American coastline. Thus, further monitoring of this and other NIS with restricted introduced ranges is strongly recommended. Finally, future studies should consider both habitat suitability and genomic data to holistically assess the potential for spread of NIS.

Materials and Methods

Study Species and Pacific-Wide Distribution.

Pyura praeputialis, previously known as P. stolonifera or P. stolonifera praeputialis (68), is a solitary ascidian (Tunicata, Ascidiacea) that forms densely packed sessile aggregates that can monopolize the intertidal and subtidal along extensive stretches of coastline. Along its native Australian habitat, P. praeputialis ranges from Cape Otway in Victoria (southeast coast) to southern Queensland (northeast coast) (Fig. 1), with a biogeographic barrier at Wilson’s Promontory separating two genetically differentiated populations of the species (35, 36). Pyura praeputialis is also present in South America as a single, isolated population in Antofagasta Bay, northern Chile (Fig. 1) (35, 37). Here, P. praeputialis achieves ecological dominance and the highest biomass per unit area reported for any intertidal species (69), reducing survival of native species by growing over them (58). It has been proposed that P. praeputialis was transported to Antofagasta Bay by shipping in the mid-late 19th century (62, 70) during the onset of the nitrate trade between Chile and Australia (71, 72). Accordingly, genetic studies show a close relation between Chilean and eastern Australian populations (35, 37).

Field Sampling and Genomic Dataset.

One hundred ninety individuals of P. praeputialis were collected along the coastline of Antofagasta Bay and from several locations throughout the species’ Australian range (Fig. 1 and SI Appendix, Table S1). Details on tissue dissection are found in SI Appendix.

Genotyping-by-sequencing (GBS) libraries were prepared at the University of Wisconsin Biotechnology Centre following Elshire et al. (73) using the ApeKI restriction enzyme. The GBS libraries were subsequently sequenced on an Illumina NovaSeq6000 sequencer. The GBS reads were assembled using ipyrad v. 0.7.30 (74) with parameters recommended for paired-end GBS data (https://ipyrad.readthedocs.io/). As no published genome of any species for the family Pyuridae is available, we used de novo assembly methods, as described in SI Appendix.

These data were used to create two SNP datasets. Firstly, a dataset of candidate loci under environmental selection was generated using two GEA analyses, bayenv2 (75) and RDA (76). The GEA methods were used to identify putative adaptive SNPs based on hypothesized associations between genotype and 13 environmental variables related to temperature, salinity, dissolved oxygen, and pH. These environmental variables were chosen as they have been shown to influence distributions of ascidians (77, 78). Details on the parameters used for the analyses are in SI Appendix. Secondly, a dataset of putatively neutral loci was generated using loci not recovered by either GEA analysis.

Comparative Population Genomics.

For each sampling site, population genomic statistics for the neutral dataset were calculated using the R package “diveRsity” v.1.9.0 (79). These included observed heterozygosity (H_O), expected heterozygosity (H_E), and inbreeding coefficient (F_IS) (80). Significance in F_IS values was inferred using bootstrapping over 10,000 permutations using the neutral dataset.

Population structure was assessed using two methods that assign individuals to predefined clusters (ADMIXTURE and DAPC), and by estimating F_ST for pairs of sites using both the neutral and candidate datasets. ADMIXTURE v.1.3 (81) was used to estimate the likelihood that an individual comes from one of a predefined number of putative sample populations (K), and a DAPC (82) was performed using the R package “adegenet” v.2.1.1. We ran the DAPC with a priori knowledge of individual populations. Pairwise population genetic differentiation was examined by calculating F_ST values following Weir and Cockerham (80) in the R package “hierfstat” v.0.04-30 (83), with significance assessed by running 10,000 permutations after correcting for multiple comparisons using Benjamini–Yekutieli false-discovery rate correction (84).

In order to reconstruct the invasion history of P. praeputialis, ABC analyses were conducted using DIYABC v.2.1.0 (85). Two sets of scenarios were tested. The aim of the first was to infer the colonization history of P. praeputialis from Australia to South America, and the second was used to estimate the effective population size of individuals that founded the population in Antofagasta Bay (see SI Appendix for a detailed explanation of these scenarios and other methodological considerations).

Species Distribution Modeling.

To determine the extent of suitable habitat for P. praeputialis across both sides of the Pacific, we used the maximum entropy method implemented in Maxent v.3.4.1 (86). We gathered spatial records by combining occurrence data from the Global Biodiversity Information Facility (https://www.gbif.com/), the Ocean Biodiversity Information System (https://www.obis.org/), and sample sites from the present study plus three previous studies (35, 36, 87). We extracted 22 ecologically relevant environmental datasets from Bio-ORACLE (88) and MARSPEC (89). To account for redundancy and the effects of collinearity among variables, we removed environmental variables that were highly correlated (|r| > 0.7) (90), retaining a total of eight variables (SI Appendix, Table S6). We produced models using all occurrence data and also models using only the occurrence points relevant to the invasion (i.e., genomics-informed). We used the R package "ENMeval" v.2.0.0 to evaluate model predictive ability and avoid overfitting, a frequently overlooked aspect of SDMs (91). Additional details are given in SI Appendix.

Data Availability

Genomic data (SNPs) have been deposited at https://doi.org/10.6084/m9.figshare.c.5428047, and R scripts are available at https://github.com/HudsonJamie/pyura-praeputialis-PNAS.