Early View
Open Access

Rapid niche shifts as drivers for the spread of a non-indigenous species under novel environmental conditions

Kathryn E. Pack,

Corresponding Author

Kathryn E. Pack

School of Ocean and Earth Science, National Oceanography Centre Southampton, University of Southampton, Southampton, UK

Marine Biological Association, Plymouth, UK


Kathryn E. Pack, School of Ocean and Earth Science, National Oceanography Centre Southampton, University of Southampton, Southampton, UK.

Email: kathrynepack@gmail.com

Search for more papers by this author
Nova Mieszkowska,

Nova Mieszkowska

Marine Biological Association, Plymouth, UK

School of Environmental Sciences, University of Liverpool, Liverpool, UK

Search for more papers by this author
Marc Rius,

Marc Rius

Centre for Ecological Genomics and Wildlife Conservation, Department of Zoology, University of Johannesburg, Auckland Park, South Africa

Centre for Advanced Studies of Blanes (CEAB, CSIC), Accés a la Cala Sant Francesc, Blanes, Spain

Search for more papers by this author
First published: 13 January 2022



Identifying niche shifts is key for forecasting future species distributions. Non-indigenous species (NIS) are one of the greatest threats to biodiversity, and understanding how niche shifts affect the spread of NIS is fundamental. Here, we modelled the native and introduced niches, as well as the potential geographical extent of a widely distributed NIS, the Pacific oyster Magallana gigas. We then tested for niche shifts in environmental space and predicted spread under contemporary climate change (CCC) conditions.




We used: (1) the two-dimensional Centroid shift, Overlap, Unfilling and Expansion (COUE) framework and (2) the n-dimensional hypervolume framework (NDH) to quantify the niches in both analogue and total environmental spaces. Niches were tested for equivalency by comparing the observed and randomized overlaps. Ensemble ecological niche models (ENMs) were then used to predict habitat suitability for the present-day and two future CCC scenarios.


The NDH framework indicated that the introduced niche of M. gigas has shifted into new environmental conditions compared to the native niche. In contrast, COUE framework implied no niche shift, but the first two dimensions only accounted for a small proportion of the overall environmental variability. Ensemble ENMs revealed suitable areas where Mgigas has yet to be recorded and predicted both a poleward expansion and a tropical contraction of suitable habitat for Mgigas by 2100.

Main conclusions

We found that Mgigas has rapidly shifted its niche in both analogue and non-analogue environmental spaces since it was first recorded as introduced species over 50 years ago. Our results suggested that niche shifts facilitate both present-day and future spread of NIS. Additionally, our study demonstrated the importance of modelling niche dynamics in multidimensional space for predicting range shifts of NIS under CCC.


Non-indigenous species (NIS) are one of the biggest threats to global biodiversity (Bax et al., 2003; Molnar et al., 2008), and understanding their potential distributions in the near future is a priority for biodiversity managers (Guisan et al., 2013; Sinclair et al., 2010). Preventing the establishment of NIS is recognized as the most efficient way of reducing their impact on ecosystems (Simberloff et al., 2013), and analyses predicting habitat suitability of NIS are fundamental for preventing NIS introductions and controlling their spread (Jiménez-Valverde et al., 2011; Leidenberger et al., 2015; Therriault & Herborg, 2008). Ecological niche models (ENMs) are popular mathematical models used to increase our understanding of suitable habitats and predicting areas that may be at risk of invasion (Mainali et al., 2015; Thuiller et al., 2005; Václavík & Meentemeyer, 2012). In addition, ENMs forecast range expansions and contractions under contemporary climate change (CCC) conditions (Elith et al., 2010; Elith & Leathwick, 2009) such as poleward range shifts (Cheung et al., 2009; Jones & Cheung, 2015; Jueterbock et al., 2016; Saupe et al., 2014). In the marine environment, range shifts are an order of magnitude faster than terrestrial ecosystems (Sorte et al., 2010), and thus investigating potential marine NIS distributions and areas at risk of invasion with CCC is of high importance. By the end of the century, the oceans will experience environmental conditions without modern analogue due to CCC (Williams & Jackson, 2007) and this is expected to lead to the creation of novel ecological communities (Pecl et al., 2017). Although an increasing number of studies using ENMs suggest that NIS will expand their ranges and/or shift their ranges poleward as a result of ocean warming (Goldsmit et al., 2018; de Rivera et al., 2011; Saupe et al., 2014), less is known about the stability of their niche space over time.

One of the main assumptions of correlative ENMs is that a species niche is stable in space and time and that the species is in equilibrium with the environment (i.e. they live in all suitable areas and are absent from unsuitable ones) (Elith et al., 2010; Hutchinson, 1957). When modelling NIS, these assumptions are unlikely to be valid due to the actual nature of species introductions. NIS are transported into a new location where the introduced species do not yet inhabit all suitable areas and are far from equilibrium (Gallien et al., 2012). This raises the question of whether to model with occurrence data from only their native range in ENMs (where the species is likely to be in equilibrium) or the entire range (Early & Sax, 2014; Guisan & Thuiller, 2005; Hudson et al., 2021). Studies using only the native range have poorly predicted habitat suitability in introduced ranges (Beaumont et al., 2009; Verbruggen et al., 2013) and thus an increasing number of ENM studies are considering both the native and introduced ranges. In addition, a growing number of studies suggest that modelling efforts can be affected by the so-called “niche shifts” (Broennimann et al., 2007; Parravicini et al., 2015; Reiss et al., 2014). Niche shifts describe the divergence in the physical and environmental requirements of a species over time and geographical space (Figure 1) due to ecological and evolutionary changes (Broennimann et al., 2007). CCC has the potential to alter species niches and identifying if NIS can undergo rapid niche shifts may help understand how the study species responds to novel climates (Guisan et al., 2014; Moran & Alexander, 2014). Coupling these investigations with ENM studies improves our forecasting accuracy and increases our understanding of the potential spread of NIS under CCC (Parravicini et al., 2015; Tingley et al., 2014). For example, ENMs that account for niche changes (by combining data from both native and introduced ranges) have led to models with higher predictive power across both ranges and under future CCC scenarios (Beaumont et al., 2009; Broennimann & Guisan, 2008; Pili et al., 2020). Quantifying niche changes, such as the degree of overlap or expansion, has recently become a rapidly expanding field of research with the development of several frameworks (Blonder et al., 2014; Broennimann et al., 2012). However, studies directly comparing different frameworks are rare (Pili et al., 2020) and further work is required to understand how these frameworks may affect the overall evaluation of niche shifts.

Details are in the caption following the image
A two-dimensional representation of a niche shift from the native (green) to introduced (red) climate spaces. The thin, outer solid lines show all the environmental conditions available in the native and introduced ranges (total environmental space). The grey area shows the environmental conditions that exist in both ranges, known as the analogue climate. The green and red thick lines depict the native and introduced niches, respectively. (a) Conditions inside the native niche but are non-analogue to the introduced range. (b) ‘Niche unfilling’ showing analogue conditions filled by the native niche but not filled by the introduced niche. (c) ‘Niche overlap’ showing analogue conditions that both the native and introduced niches occupy. (d) ‘Niche expansion’ showing conditions in the introduced niche not occupied by the native niche. (e) Conditions inside the introduced niche but are non-analogue to the native range. The quantification of b, c, and d imply the level of niche shift (or niche conservatism) between the native and introduced niches. Note that the introduced niche is larger than the native niche, as expected considering the extensive introduced range of the species included in this study. Figure redrawn from Guisan et al. (2014)

Quantifying niche dynamics across different geographic ranges requires a distinction between analogue and non-analogue environments (Figure 1) (Guisan et al., 2014). If the introduced niche shifts into non-analogue climates (i.e. where environmental conditions are not found in the native range), it cannot be directly assumed that the niche has shifted as it could simply be a factor of these environmental conditions being unavailable in the native range (Guisan et al., 2014; Mandle et al., 2011). In analogue climates (i.e. where the same environmental conditions are available in both native and introduced ranges) niche changes demonstrate “true niche shifts” as the introduced individuals inhabit environments that are available in both ranges, but the native do not inhabit (Figure 1). Nonetheless, colonization of non-analogue climates from the introduced range still provides crucial information on the potential tolerance of NIS to novel climates and has important implications for future management strategies (Early & Sax, 2014). The majority of studies modelling niche shifts do not distinguish between non-analogue and analogue climates and, therefore, may just be identifying environmental conditions that are unavailable in one range (Guisan et al., 2014).

Here, we first assessed the potential for a highly successful NIS, the Pacific oyster Magallana gigas (Thunberg, 1793), to shift its niche in total environmental (non-analogue and analogue) and analogue climates using different frameworks for quantifying niche shifts. Mgigas has wild populations in more than 17 countries mainly due to the species artificial introductions for aquaculture purposes (Herbert et al., 2016). Field observations have shown that warming over the last few decades has facilitated the spread of Mgigas along the coastlines of Europe and North America (Diederich et al., 2005; Valdez & Ruesink, 2017; Wrange et al., 2010). Laboratory experiments have also shown that M. gigas can tolerate a wide range of environmental conditions (Pack et al., 2021). Therefore, studying whether Mgigas has undergone rapid niche shifts during its invasion may further unravel its ability to thrive under novel CCC conditions. In this study we used ENMS with data from both native and introduced ranges to forecast if, under continued warming scenarios predicted by the end of the century, Mgigas distributions will continue to extend poleward as new locations become suitable. We hypothesized that: (1) quantifying the degree of niche overlap in non-analogue and analogue environmental space would show the introduced niche of Mgigas has expanded compared to the native niche, (2) different niche dynamic frameworks would show similar results in terms of overlap and expansion of Mgigas niches, and (3) ensemble ENMs would forecast a poleward expansion of suitable geographic areas for Mgigas by the end of the century.


2.1 Environmental layers

Present-day environmental layers for the surface ocean were obtained from Bio-ORACLE (version 2.0) at 5 arcminute (9.2 km2 at the equator) resolution (Assis et al., 2018; Tyberghein et al., 2012). This database contains global variables collected from satellite sensors and interpolated in situ measurements averaged across 2000–2014 (Assis et al., 2018). For modelling in both environmental and geographical space, we used five environmental variables that are known to be biologically relevant to sessile filter-feeders and were not highly correlated when tested for collinearity (Pearson's correlation coefficient of r > 0.75). The five variables were temperature, salinity, pH, chlorophyll concentration and calcite concentration (Table S1.1).

To estimate future climatic conditions, a bias correction method using predicted environmental conditions from the HadGEM-ES general circulation model from phase five of the Coupled Model Intercomparison Project (Met Office Hadley Centre, 2020) was applied to the Bio-ORACLE present-day layers. Firstly, simulated monthly values for each variable were extracted from the HadGEM-ES CMIP5 for 2006–2014 and 2090–2100 and the overall mean for each period was calculated per 1° × 1° grid cell for two scenarios based on the IPCC “Representative Concentration Pathways”: RCP4.5 (stabilization scenario) and RCP8.5 (“business as usual” scenario). The change factor (delta) between the simulated present-day and future variables for each scenario were calculated for each grid cell and the delta applied to the Bio-ORACLE variables (aggregated to 1° × 1°) to estimate future conditions (Navarro-Racines et al., 2020). Calcite was removed from present and future projections as a measure of end of the century as calcite concentration was not available from the HadGEM-ES model.

2.2 Species records

Global presence-only occurrence records for Mgigas were downloaded from the Global Biodiversity Information Facility (GBIF, https://www.gbif.org/species/7820753) and contrasted with literature data for verification. The data were cleaned to remove records of Mgigas that were unverified, recorded as dead/empty shells, “washed ashore,” and records which were inland to obtain a dataset containing georeferenced records or collected museum records. In total there were 1864 unique presence records. For the comparison of niches in environmental space, occurrence points were spatially aggregated to one per 0.08° × 0.08° (5 × 5 arcminute) resolution in line with higher resolution present-day environmental grid cells, leaving 663 records. For habitat suitability modelling, occurrence points were spatially aggregated to one per 1° × 1° environmental grid cell, in line with the end-of-the-century environmental data, to reduce spatial clustering and the effects of sampling bias, leaving 231 records (Figure S1.1).

2.3 Comparison of niches in environmental space

Species records were split into two data sets containing native (N = 54) and introduced (N = 609) records and present-day environmental data for each of the ranges were extracted. Firstly, the environmental conditions that the introduced and native occurrence points occupy were compared by plotting their kernel densities distributions with each environmental variable. Then, two popular modelling methods were used to quantify the niches in environmental space; the widely used Centroid shift, Overlap, Unfilling and Expansion (COUE) framework (Broennimann et al., 2012) and the n-Dimensional Hypervolume framework (NDH) (Blonder et al., 2014), which measure the degree of overlap and environment equivalency between niches.

2.3.1 The centroid shift, overlap, unfilling and expansion framework

Using the COUE framework the niches were compared in two-dimensional space using Schoener's D metric of niche overlap (Broennimann et al., 2012). Principal component (PC) analysis was used on the extracted environmental data from the native and introduced ranges. The first two PCs accounted for 57% of the variability. The environmental space was divided into 100 grid cells and the density of occurrences within the environmental space was estimated using a kernel density estimator (Broennimann et al., 2012; Parravicini et al., 2015). Schoener's D metric quantified the degree of overlap of the two niches ranging between 0 (no overlap) and +1 (complete overlap) (Broennimann et al., 2012; Warren et al., 2008).

A test of niche equivalency was performed to test whether the observed overlap (D) was significantly different to a null distribution of 1000 generated D metrics by randomly reallocating the occurrences of both niches into two datasets (Broennimann et al., 2012; Warren et al., 2008). The null hypothesis that the niches are identical was rejected if the observed D was below the 5th percentile of the null distribution. A test for niche similarity was performed to assess if the observed D was significantly different to a null distribution of 1000 generated D metrics when accounting for geographic availability of environmental conditions (by randomly distributing one niche over its background whilst the other is unchanged). The null hypothesis that the niches are dissimilar was rejected if the observed D was greater than the 95th percentile of the null distribution (Broennimann et al., 2012; Warren et al., 2008).

In addition, the COUE measures of niche stability (i.e. the proportion of environments within the introduced niche shared with the native niche), niche expansion (i.e. the proportion of environments of the introduced niche that do not intersect with the native niche), and niche unfilling (the proportion of environments not currently filled by the invasive niche) were reported (Guisan et al., 2014). Niche metrics described above were calculated in R version 4.0.3 (R Core Team, 2019) with the ecospat package version 3.1 (Di Cola et al., 2017).

2.3.2 The n-Dimensional Hypervolume framework

As noted above, the first two PC in the COUE framework accounted for only 57% of the total variability in the five environmental variables; therefore, important information on the differences between the niches may not have been satisfactorily captured in these first two components. This may be expected since the PC method does not make use of any information on groupings in the data such as the native and introduced populations.

The NDH framework was used to assess differences in the niches within a multidimensional environmental space (Blonder et al., 2014; Hutchinson, 1957). The introduced and native niches were quantified in five dimensions, accounting for 100% of the variability in the environmental data. The five PCs of the original five environmental variables were used to allow a direct comparison with the results of the COUE framework, which uses only the first two PCs. Multidimensional hypervolumes for the native and introduced niches were generated using the Gaussian kernel density estimation method (Blonder et al., 2014). Hypervolume niche metrics and analyses were conducted using the hypervolume R package (Blonder et al., 2014).

To assess the differences between the niches, Jaccard's index and Sørensen-Dice index were calculated by:

Sørensen-Dice urn:x-wiley:13669516:media:ddi13471:ddi13471-math-0001

Jaccard urn:x-wiley:13669516:media:ddi13471:ddi13471-math-0002

where A and B are the two hypervolumes, ∩ is the intersection between the two hypervolumes and ∪ is the union of the two hypervolumes (Blonder et al., 2014). Similar to the COUE equivalency test, the Sørensen-Dice and Jaccard indices were compared to a null distribution from 1000 iterations to assess significance. Jaccard and Sørensen-Dice metrics were also calculated for the first two PCs (similar to the COUE framework) to allow for a more direct comparison between the two frameworks.

2.3.3 Analogue environment

To better understand if changes in the introduced niche were as a result of a “true niche shift”, the two niches were modelled as above but in analogue environments, which were extracted using the ecospat package in R. Based on Mesgaran et al. (2014), the package uses Mahalanobis distance to detect covariate correlations between the univariate range of covariates. The presence records and associated climates within the analogue environment were then modelled in environmental space.

2.3.4 Partial least squares

The first two PCs may not contain information pertinent to the differences between native and introduced niches, especially when the first two components only explain a relatively low percentage of the total variation. This implies that visualizations based on the first two PCs may be uninformative. Partial least squares (PLS), used as a discriminant analysis technique, may better illustrate population differences as it maximizes the covariance between the environmental data from the two niches. As the PLS approach can be influenced by sample sizes, the datasets were balanced by clustering the introduced records using a k-means approach into 54 clusters (i.e. number of records in the native range) and the centroids of each cluster formed into a new dataset of 54 records. This clustering approach ensured that the new dataset best covered the space of the introduced niche. The new factors and scores from the PLS were extracted (equivalent to PCA).

2.4 Habitat suitability modelling

As a broad geographical extent allows for a closer approximation of the fundamental niche (Broennimann & Guisan, 2008; Broennimann et al., 2007; Phillips et al., 2006), ENMs were calibrated using occurrence records from both the native and introduced ranges. We then used these records to predict the current and future habitat suitability of Mgigas around the globe.

An ensemble model of four widely used presence-only models, BIOCLIM, Mahalanobis distance, Maxent and Boosted Regression Trees (BRT), was used to estimate habitat suitability of Mgigas globally. Each individual model was run using a bootstrap with replacement method, replicated 100 times and the median model determined. Habitat suitability in the ensemble model was calculated as the mean probability per grid cell weighted by their performance (area under the curve, see Section 2.4.1). Habitat suitability was modelled for the present-day using all five present-day environmental variables and for two end-of-the-century scenarios based on RCP4.5 and RCP8.5 projections. The end-of-the-century models were trained on environmental variables and occurrence records in the present-day then projected into conditions for each scenario.

2.4.1 Model validation and variable importance

To evaluate model performance, the area under the curve (AUC) of the receiver operating characteristic (ROC) and true skill statistic (TSS) were calculated based on 20 model runs with a randomly sampled training and test data set of 75% and 25% of the presence records, respectively.

The importance of each of the five variables in the present-day model was examined using a leave-one-out technique. BIOCLIM, Mahalanobis distance, Maxent and BRT were run and evaluated based on models with only four of the five environmental variables and the mean AUC and TSS for each leave-one-out model was calculated based on 20 randomly sampled training and test datasets. The changes in the AUC and TSS between the models were then compared.

To assess the robustness of the habitat suitability model results, the ensemble models were rerun using different levels of presence records by applying a set of thresholds. For Mgigas to be considered present in a given 1° × 1° grid cell, either 3, 5, or 7 records were required. These three threshold levels resulted in three different presence datasets, each of which was used to predict habitat suitability under the present-day and end-of-the-century RCP8.5 scenario. Each model was run 100 times using a bootstrapping with replacement method and the ensemble models calculated using the same method as the original model. The model predictions were then compared to the original model.

2.4.2 Latitudinal shift in habitat suitability

The average poleward shift of suitable habitat for Mgigas was predicted between the present-day ensemble model and the two end-of-the-century ensemble models. Firstly, both ensemble models were split geographically into northern and southern hemispheres to account for the differences in the direction of movement above and below the equator. Then, the latitudinal centroids urn:x-wiley:13669516:media:ddi13471:ddi13471-math-0003, weighted by habitat suitability, were calculated using
where urn:x-wiley:13669516:media:ddi13471:ddi13471-math-0005 is the latitude at the centre of the spatial cell urn:x-wiley:13669516:media:ddi13471:ddi13471-math-0006, urn:x-wiley:13669516:media:ddi13471:ddi13471-math-0007 is the habitat suitability in the cell, and urn:x-wiley:13669516:media:ddi13471:ddi13471-math-0008 is the total number of cells (Cheung et al., 2009; Jones & Cheung, 2014). The difference between the latitudinal centroid was then calculated and converted to kilometres using:
where urn:x-wiley:13669516:media:ddi13471:ddi13471-math-0010 and urn:x-wiley:13669516:media:ddi13471:ddi13471-math-0011 are the latitudinal centroids for present day and future, respectively, and 6378.2 is the approximate diameter of the Earth in kilometres (Cheung et al., 2009; Jones & Cheung, 2014). Habitat suitability was averaged per 1° latitude for the present day and end-of-the-century models and the difference was calculated.


3.1 Niche overlap

3.1.1 COUE and NDH niche shifts

The univariate density curves showed that the biggest difference between the climatic niches occurred for temperature where the optimum temperature in the native niche was 19.6°C and the optimum in the introduced was 12.1°C (Figure 2a). The introduced niche occupied a similar but broader set of conditions than the native niche for salinity, pH and calcite (Figure 2b,c,e). The introduced and native niches occupied a similar range of chlorophyll concentrations (Figure 2d).

Details are in the caption following the image
The kernel density of native and introduced range occurrences of Magallana gigas for the environmental variables used in the ensemble models. The five environmental variables include temperature (a), salinity (b), pH (c) chlorophyll (d), and calcite (e)

The COUE framework suggested that the introduced niche covered a broader set of environmental conditions than the native niche (Figure 3a). Schoener's D metric indicated that niche overlap was moderately low (D = 0.12) and the climatic niche of introduced Mgigas has expanded by approximately 30% compared to the native (Figure 3a) with niche stability at 70% and unfilling niche at 0%. The null hypothesis of niche equivalency was accepted (p = 0.21, Figure 3b) implying the niches were the equivalent. The environments occupied by the introduced niche were more similar to the native niche than expected by chance (p = 0.032) (Figure 3c). These results were expected as the native niche was completely overlapped by the introduced niche in Figure 3a. The environments occupied by the native niche were not more similar than expected by chance to the introduced niche (p = 0.10) (Figure 3d), reflecting the expansion in the introduced niche.

Details are in the caption following the image
Principal components niche comparisons in two-dimensional space using the COUE framework in the total environmental space (non-analogue and analogue) (a–d) and analogue space (e–h) of Magallana gigas. The niches of M. gigas in the native and introduced ranges and the overlap in environmental space (a, e). The solid and dashed lines represent 100% and 75% of the available environmental space in the native (green) and introduced (red) ranges and shading indicates the density of occurrences. Blue indicates overlapping of the niches, green indicates unfilling in the native niche and red indicates expansion in the introduced range. Histograms showing the null distribution of D from 1000 iterations compared to the observed Schoener D metric (red diamond) for niche equivalency (b, f), and similarity for the introduced to native niche (c, g) and native to introduced niche (d, h)

Jaccard and Sørensen-Dice indices in two-dimensional space were moderately high in two-dimensional space at 0.34 and 0.51, respectively (Table 1), reflecting the overlap observed from the COUE framework. Only 14% of the native niche was unique and 64% of the introduced niche was unique. These results imply that Mgigas in the introduced range have colonized nearly all the available climate conditions occupied by the native niche and has expanded into novel climates.

TABLE 1. The metrics for determining the proportion of the niches that are unique, and the Jaccard and Sørensen-Dice indices for the union between the native and introduced niches of Magallana gigas in environmental space
Total environment Analogue environment
2 dimensions 5 dimensions 2 dimensions 5 dimensions
Fraction of unique niche outside of the overlap (native/introduced) (%) 14% / 64% 66% / 83% 50% / 75% 86% / 87%
Jaccard index 0.34 0.13 0.20 0.07
Sørensen-Dice index 0.51 0.23 0.33 0.13


  • The metrics were quantified in two-dimensional and five-dimensional principal components space for the total environment (combined non-analogue and analogue environments) and the analogue space.

Similar to the COUE framework, the five-dimensional NDH showed that the introduced niche covers a broader range of environmental conditions than the native niche (Figure 4a). The first plot of the hypervolume (PC1 vs PC2, Figure 4a) is synonymous to the COUE plot (Figure 3a). There was low overlap at the 95% percentile probability boundary of the hypervolumes with 0.23 and 0.13 from the Sørensen-Dice and Jaccard indices, respectively (Table 1). Contrary to the two-dimensional metrics, 66% of the native hypervolume was unique and 83% of the introduced hypervolume was unique (Table 1). The null hypothesis of niche equivalency was rejected (Sørensen-Dice: = 0.003, Jaccard: = 0.003, Figure 4b,c), thus the niches were not more equivalent to each other than expected by chance in a five-dimensional space.

Details are in the caption following the image
(a) Five-dimensional hypervolume in pairwise, two-dimensional space comparing the niches of the native (blue) and introduced (red) ranges of Magallana gigas in the total environmental space. The filled circles indicate the centroid of the hypervolume and the small, filled dots represent true species presence records. The native points overlay introduced points. The observed values from the Sørensen-Dice (b) and Jaccard indices (c) (red diamond) are compared to a histogram of the null distribution of simulated values using a randomisation method with 1000 iterations

3.1.2 Niche shifts in analogue environments

Isolating the analogue environment showed that approximately 61.7% of the introduced occurrences (376 of the total 609) exist in areas with environments analogous to the native range, with the remaining 38.3% living in environments deemed non-analogue. From the COUE framework, the Schoener index implied moderate overlap, D = 0.31, with an expansion in the introduced niche of 17.8%, niche stability of 82.2% and niche unfilling of 24.1% (Figure 3e). The null hypothesis of niche equivalency was accepted (= 0.39, Figure 3f) implying the niches were equivalent. The environments were also more similar than expected by chance for both niches (Figure 3g,h). The two-dimensional Jaccard and Sørensen-Dice indices for the analogue space were 0.20 and 0.33, respectively with 50% of the native niche and 75% of the introduced niche being unique (Table 1).

In analogue space, a clear separation in the native and introduced hypervolumes can be observed between the niches between PC2 and PC3 (Figure 5a). The Sørensen-Dice and Jaccard indices for the five-dimensional hypervolume indicated very low intersection at 0.07 and 0.13, respectively, and a high proportion of the native and introduced niches were unique, 86% and 87%, respectively (Table 1). The niches were considered not equivalent as the observed Sørensen-Dice and Jaccard indices were less than 95% of simulated values in the null distribution (Sørensen-Dice: < 0.001, Jaccard: < 0.001, Figure 5b,c).

Details are in the caption following the image
(a) Five-dimensional hypervolume in pairwise, two-dimensional space comparing the niches of the native (blue) and introduced (red) ranges of Magallana gigas in an analogue environmental space. The large, filled circles indicate the centroid of the hypervolume and the small, filled dots represent true species presence records. The observed values from the Sørensen-Dice (b) and Jaccard (c) indices in the analogue environmental space (red diamond) are compared to a histogram of the null distribution of simulated values using a randomisation method with 1000 iterations

3.1.3 Partial least squares

Partial least squares were consistent with the hypervolume statistics, showing a clearer separation between the native and introduced niches within the first two PLS factor (Figure S1.2). The first PLS was predominately based on temperature in contrast with calcite, hence temperature was largely responsible for the separation between the native and introduced niches (Table S1.2). The pairwise PC plots misleadingly suggested a high level of overlap with the COUE framework implying a complete overlap of the native by the introduced niches (Figure 3). This was reflected in the eigenvectors that showed that none of the variables had a strong influence on the axes (Table S1.2), thus not representing the difference between the native and introduced niches.

3.2 Ecological niche modelling

Moderate to high habitat suitability in temperate regions of the northwest and northeast Pacific, north Atlantic and Australasia were predicted for the present day, with the northern European coast and UK predicted to have the most suitable habitat for Mgigas, highlighting areas where dense populations can currently exist (Figure 6a). The model predicted areas of moderate habitat suitability in areas where few presence records have been recorded, such as the coasts of South America and South Africa and east coast of North America. The final ensemble model had high predictive ability with an AUC of 0.92 and the TSS of 0.74. Standard deviations implied a low level of variation of habitat suitability (Figure S1.3).

Details are in the caption following the image
Ensemble maps for habitat suitability for Magallana gigas along the world's coastline for the present day (a) and the end of the century based on RCP4.5 and RCP8.5 scenarios (b, c), respectively. The ensemble predictions are based on the median BIOCLIM, Mahalanobis distance, Maxent and BRT predictions

Overall, habitat suitability decreased for both scenarios predicted by the end of the century compared to the present-day ensemble model (Figure 6b,c). Habitat suitability for Mgigas was predicted to increase at higher latitudes such as the west coast of Canada and Alaska, northwest Europe and northeast Asia by the end of the century in both scenarios (Figure 6b,c). A decline in habitat suitability was predicted in some regions, such as along the coasts of Australasia, southern Africa and south America, particularly under the ‘business as usual’ scenario (RCP8.5) (Figure 6c).

Under both RCP scenarios, the two envelope models, BIOCLIM and Mahalanobis distance, predicted low to no suitable habitat by the end of the century (Figure S1.4). Maxent and BRT models were, therefore, responsible for the predicted habitat suitability in the ensemble models. All four models had good accuracy (Figure S1.5). The leave-one-out method for variable importance showed consistent results across all four models. The removal of temperature had the greatest effect on the AUC and TSS. Temperature was, therefore, the most important predictor of habitat suitability for M. gigas. The other four variables had minimal impact on the model accuracy when omitted (Figure S1.5).

Both the northern and southern hemisphere latitudinal centroids of habitat suitability shifted polewards by the end of the century in both IPCC scenarios (Table 2). As expected, the predicted shift in the latitudinal centroids was more extreme for RCP8.5 (924.4 and 785.4 km) than for RCP4.5 (501.7 and 203.5 km). The latitudinal differences between the present-day and future scenarios showed that habitat suitability decreased in lower temperate latitudes but increased in the higher latitudes by approximately 10%–25% (Figure S1.6).

Thresholding of the occurrence data showed consensus with the original models (see Appendix S2). All the models predicted similar habitat suitability globally, however suitability and predicted geographical extent marginally decreased where fewer presence locations were included at the range edges, as might be expected (Figure S2.1 for comparisons). In addition, suitable geographic regions were predicted even when the thresholds removed presence data in these regions, such as the coastline of South Africa and South America (Figure S2.1, S2.2). Predicted latitudinal shift between the present-day and RCP8.5 model was also similar across the models (Table S2.1). Overall, the similarities between these model predictions suggested that the original model results remained robust.

TABLE 2. The predicted latitudinal shift in Magallana gigas under RCP4.5 and RCP8.5 climate change scenarios
Hemisphere Present-day centroid (decimal degrees) Future centroid (decimal degrees) Difference (decimal degrees)



RCP4.5 45.5 34.6 50.0 36.6 4.5 2.0 501.7 203.5
RCP8.5 53.8 41.7 8.3 7.1 924.4 785.4


  • The latitudinal centroids for both the present-day and end of the century ensemble models were determined for the northern (N) and southern (S) hemispheres. The predicted future latitudinal shifts were then calculated as the difference between the present and future centroids and expressed in kilometres.


In environmental space, we found that Mgigas in the introduced range has shifted their niche relative to the native niche in both analogue and non-analogue environments. These results indicated that Mgigas can thrive in novel environments, suggesting the presence of mechanisms that may facilitate present-day and future spread. We found partially inconsistent results from examining niche shifts using two popular frameworks, as the five-dimensional NDH framework showed shifts in the introduced niche and the two-dimensional COUE framework implied niche conservatism. However, the first two dimensions used in the COUE framework only accounted for a small proportion of the overall environmental variability. Thus, this study highlights the need for careful interpretation of the results when comparing niches in two-dimensional space, particularly where the two PCs explain a low amount of the variability in the environmental data. Areas of suitable climates for Mgigas were generated for both the present day and end of the century, with results predicting an increase in suitable habitat towards the poles and a contraction towards the tropics under two forecasted CCC scenarios by 2100. This agrees with previous ENM studies that showed that towards the end of the century, several NIS are predicted to undergo poleward range shifts (Goldsmit et al., 2018; Lowen & DiBacco, 2017; de Rivera et al., 2011) due to CCC.

The results from this study support previous findings showing that NIS across a range of habitats and taxa are undergoing rapid niche shifts in their introduced ranges (Li et al., 2014; Parravicini et al., 2015; Pili et al., 2020; Tingley et al., 2014; Torres et al., 2018; Zhang et al., 2020). Parravicini et al. (2015) demonstrated that 33% of the invasive tropical fishes studied in the Mediterranean Sea showed expansions in their introduced niches and concluded that ENMs that do not account for niche shifts may underestimate their potential spread. In freshwater ecosystems, up to 90% of the 22 invertebrate species exhibited significant changes in their introduced niche compared to their native niche (Torres et al., 2018). Identifying niche shifts gives an indication of the potential for a species to adapt and spread into novel climates both currently and in the future, which may ultimately increase NIS success with CCC.

Our modelling in analogue climates indicated that a “true” niche shift has occurred in the introduced range of Mgigas over a relatively short time since its global introduction for aquaculture purposes. Several factors may contribute to an observed shift in the realized niche of the introduced range (Alexander & Edwards, 2010; Pearman et al., 2008): expansion in the introduced niche or unfilling in the native niche may be a result of rapid evolution, a change in ecological interactions (for example, competition and predation), or changes in dispersal capabilities (Alexander & Edwards, 2010; Atwater et al., 2018; Rödder & Lötters, 2009; Sotka et al., 2018). Rapid evolutionary changes, such as adaptation during early life-history stages to novel climates (Hoffmann & Sgrò, 2011; Whitney & Gabler, 2008), hybridization of previously isolated native genotypes (Rius & Turon, 2020), and broadening environmental tolerances (Davidson et al., 2011; Tepolt & Somero, 2014) are commonly observed in studies of NIS. Mgigas has been shown to be genetically diverse in both native and introduced populations (Boom et al., 1994; English et al., 2000) and exhibit high phenotypic plasticity in terms of growth and survival across a wide range of environmental conditions (Hamdoun et al., 2003; Li et al., 2018; Taris et al., 2006). For example, growth in adult oysters occurs in temperatures between 10–40°C and salinities of 10–30, and can spawn across a wide range of conditions (Troost, 2010). This broad environmental tolerance suggests that a niche shift in analogue climates could be due to an increased availability of suitable habitats due to, for example, reduced levels of competition or predation within the introduced range. This is further supported by the high number of Mgigas inhabiting non-analogue climates which could also be a result of habitat availability or rapid evolution. Future studies should couple their investigations with experiments investigating variability in physiological tolerances between native and introduced ranges, and to gain insights into the underlying causes behind niche shifts in NIS.

Habitat suitability models can be useful for guiding NIS monitoring programmes that can aid conservation and management efforts (Crall et al., 2013). In this work, the present-day models suggest there are climatically suitable areas of the globe where Mgigas have not yet been recorded (e.g. South America and South Africa). Similarly, modelling in environmental space highlighted areas of unfilling in the native niche. These results are likely a factor of either Mgigas not having been introduced, not yet spread to these locations from their points of introduction, or a lack of survey effort in these areas. Occurrence records for marine invasive species tend to be opportunistic or incidental (Guillera-Arroita et al., 2015; Phillips et al., 2009), a widespread problem is the lack of true presence and absence records, which may directly affect our perception of species niches (Elith et al., 2006). The present habitat suitability maps for Mgigas can, however, be used to highlight areas at potential risk to invasion and guide targeted surveys, particularly in under-represented areas, which would ultimately increase the number of records and make for more robust model predictions (Feeley & Silman, 2011; Ruiz & Hewitt, 2002).

The present and end-of-the-century ENMs predicted suitable geographic areas vulnerable to future invasion. However, this relies on whether the species could be artificially introduced or successfully spread to these areas. Factors such as natural (propagules/larvae) or artificial (shipping and aquaculture) dispersal and species interactions are rarely incorporated into ENMs (Wisz et al., 2013) but are important factors to understand NIS’ absence from areas of its fundamental niche (Araújo & Guisan, 2006; Briscoe et al., 2019; Hastings et al., 2005). There have been advances in hybridizing ENMs that incorporate dispersal models that have shown an increased accuracy in predicting NIS distributions (Václavík & Meentemeyer, 2009; Wisz et al., 2013). However, whilst incorporating biological interactions has made recent progress (Araújo et al., 2014; Araújo & Luoto, 2007), challenges remain in determining which interactions are important for structuring communities and having sufficient data on geographical scales. Predictions of Mgigas distributions would highly benefit from inclusion of these biotic factors to determine their realized niche at regional and local scales.

The COUE and NDH frameworks used for examining native and introduced niches showed contrasting results. The hypervolume for both the total and analogue environmental space supported the likelihood that introduced Mgigas have undergone a niche shift. These results were inconsistent with the COUE framework, which found niches to be equivalent and similar, implying niche conservatism. The five-dimensional hypervolumes showed that there were large unique fractions of both niches occurring outside their overlap (Figures 4a, 5a). It is likely, however, that the COUE framework is limited by comparing niches in only two dimensions. Other studies comparing both of these popular frameworks have similarly shown differences in the results of niche shifts (Pili et al., 2020). Conversely, Tingley et al. (2014) found both methods provided similar results; however, the first two PCs in their study accounted for 88.9% of the variability in the data. It is therefore unlikely that modelling in higher dimensions would have changed their conclusions. Studies investigating niche shifts commonly find low correlations among the studied environmental variables. The first two PCs do not necessarily contain information relevant to the differences between niches, as PC analysis only attempts to explain variability in the environmental data. The use of the NDH framework allowed us to account for 100% of the variability in the environmental data, therefore providing a more complete understanding of the differences between the native and introduced niches.

Sørensen-Dice and Jaccard indices predicted low niche overlap in Mgigas niches; however, the two-dimensional PC plots (Figure 3a,e) did not represent this difference. The pairwise PC plots misleadingly suggested a high level of overlap. This provided evidence that the PCs did not in themselves contain the relevant information to enable a visualization of the niche differences, thus niche overlap analyses using only the first two PCs should be interpreted with caution. For visualizing the data, the PLS approach found new factors (analogous to PCs) specifically to explain species differences using the first two PLS factors and provided an informative 2D projection. Two PCs allowed for a simpler interpretation of high-dimensional data; however, in cases where the first two PCs accounted for a low amount of the variability in the dataset, higher-dimensional techniques or PLS should be explored.


This study revealed rapid niche shifts in a highly successful marine NIS in its introduced range. Niche divergence and the broad environmental tolerances shown by M. gigas highlighted its ability to establish in novel climates and may facilitate its spread under CCC conditions. From comparing the popular COUE and NDH frameworks, we highlighted that modelling niches in more than two dimensions in environmental space should be considered for analysing niche overlap, especially when low dimensions do not describe a high amount of the variability in environmental variables. In addition, we predicted a poleward expansion and tropical contraction of suitable habitat by the year 2100 under two RCP scenarios. Finally, exploring niche shifts of highly successful NIS advances our understanding on the establishment and spread of particular NIS, but the underlying causes of niche shifts require further investigation before they can be fully integrated into ecological niche models.


This work was supported by the Natural Environment Research Council, United Kingdom [Grant number NE/L002531/1].


    The Authors declare that there is no conflict of interest.


    The peer review history for this article is available at https://publons.com/publon/10.1111/ddi.13471.


    The peer review history for this article is available at https://publons.com/publon/10.1111/ddi.13471.


    Kathryn E. Pack is a marine ecologist primarily investigating the effects of contemporary climate change and ocean acidification on non-indigenous species physiology and distribution.

    Nova Mieszkowska is a physiological ecologist who runs the MarClim intertidal time series and links omics to ecosystem-level research on the impacts of climate change and ocean acidification on marine species.

    Marc Rius uses cutting-edge analytical and high-throughput sequencing techniques to conduct studies on biogeography, community ecology, population genetics and biodiversity conservation, with a special focus on assessing how human activities reshape species ranges.

    Author contributions: Kathryn E. Pack, Conceptualization, Methodology, Formal analysis, Investigation, Visualization, Writing - original draft, Writing - review & editing. Nova Mieszkowska, Funding acquisition, Writing - review & editing. Marc Rius: Funding acquisition, Writing - review & editing.

      The full text of this article hosted at iucr.org is unavailable due to technical difficulties.