Global gradients in the distribution of animal polyploids

My results reveal the clear existence of a latitudinal polyploid gradient in animals. Of the 471 polyploid species identified in this study 80.5 % have a mean latitudinal range outside the Tropics of Cancer and Capricorn. This disparity is most evident in insects where 97% of polyploid species occur outside of the tropics but is also apparent in amphibians and ray-finned fishes where the frequency of temperate species is far greater in polyploids than diploids (50.9 vs. 21.0% in Amphibia and 78.7 vs. 42.2% in Actinopterygii) (Fig. 2). Additionally, absolute latitude was a significant (P < 0.05) variable in phylogenetic ANOVAs, with polyploid species occurring at higher latitudes than diploids across all three clades (Fig. 3).

Analysis of the world’s biomes reveals similar patterns. Temperate and polar biomes including tundra, taiga, temperate broadleaf, and temperate coniferous forests all had greater relative frequencies of polyploids than diploids in each clade. By contrast, biomes such as tropical and subtropical dry or moist broadleaf forests and mangroves had reduced frequencies (SI Appendix, Fig. S1). For example, diploid amphibian species occur in tropical and subtropical dry broadleaf forests at a frequency nine times higher than that for polyploid species. At finer scale resolution, on the level of ecoregions, fewer similarities exist between the three clades however the general pattern remains. Ecoregions within the 99th percentile of polyploid frequency include the Scandinavian and Russian taiga for amphibians, upper Yukon for ray-finned fishes, and Arctic desert for insects. As with tropical biomes, species-rich ecoregions have relatively few polyploids. In amphibians, the three ecoregions with the highest species richness (all Andean montane forests) have no recorded polyploid occurrences whatsoever.

The tendency for polyploid species to occur more frequently at higher latitudes has long been observed in plants as well as several branchiopod species (37–42), though to my knowledge has never been reported for the clades considered by this study nor has it been demonstrated as a generalized, statistically significant trend in animals. Several hypotheses have been proposed to explain this phenomenon. First, environmental conditions associated with higher latitudes may simply create more polyploids. Cold shocks are a common way to induce WGD a in a wide variety of species (27, 29, 43–45). Additionally, heavily glaciated regions facilitate hybridization through secondary contact, potentially creating allopolyploids (polyploids formed through hybridization) (46, 47). In addition to being conducive to the formation of polyploids, high-latitude environments may also be selectively favorable to the survival and maintenance of polyploid lineages. Many polyploids have broader ecological tolerances than their diploid counterparts and are predicted to be more adaptable to less hospitable or newer environments (48–51). This is because the increased genome size following WGD affords polyploids more genetic variation and a higher average mutation rate per gene (22, 52). Multiple gene copies also create opportunity for adaptation, as each gene now has a redundant copy that is free to specialize or acquire new function (4, 20, 53).

Last, it is well established that higher latitude environments exhibit reduced species richness, a near-ubiquitous pattern known as the latitudinal diversity gradient (54). This reduced richness may selectively favor polyploids. Newly formed polyploids are often at risk of extinction from surrounding diploid species, either through direct competition (neopolyploids often suffer from decreased fitness (20–22, 43)) or inviable cross-ploidy mating (a phenomenon known at the minority cytotype exclusion principle (25, 55)). Polyploids may therefore be more likely to establish themselves in species poor areas with less competition and greater prezygotic reproductive barriers between themselves and surrounding diploids (2, 56, 57). A recent study in plants confirmed the existence of a latitudinal polyploid gradient and attributed it in large part to these factors (33).

Geographic patterns of animal polyploids reveal a tendency for polyploidy incidence to increase with latitude. Several hypotheses may explain this trend, as many variables that may contribute to conditions that are favorable for the formation and/or survival of polyploid lineages also covary with latitude. To assess and distinguish between these hypotheses, I collected environmental data from each occurrence in the dataset, averaged across species. Comparisons between diploid and polyploid species revealed significant trends shared across clades. Of the 14 potential explanatory variables, 13 were found to be significant (Bonferroni-corrected P < 0.05) in at least one clade and six were significant in all three clades (Figs. 2 and 3). These six include all four variables pertaining to temperature in addition to glaciation and absolute latitude itself. Temperature seasonality had the largest absolute Cohen’s d across clades, followed by change in temperature since the LGM and mean annual temperature (Fig. 3). These results suggest polyploids tend to occupy areas that are colder and have greater annual temperature fluctuations than diploids. This result is unsurprising, given that cold temperatures are known to contribute to unreduced gamete formation, which can lead to polyploidy. Indeed, cold shocks are an established method to produce artificial polyploids in aquaculture and has been demonstrated in ray-finned fishes, frogs, and branchiopods (27, 28, 44, 58). However, polyploids are also more common in those areas which have undergone large temperature shifts since the LGM (positive for mean annual temperature, negative for temperature seasonality), which suggests an adaptive response. In addition to the adaptive advantages outlined above, polyploids may be particularly well suited to cold climates. The effects of polyploidy are varied; however, one consequence that appears relatively consistent in animals is an increase in cell and/or body size, both of which are predicted to be advantageous in colder environments (59, 60).

Overall, temperature dynamics and changing climates appear to be the most dramatic differences between polyploid and diploid distributions. However, my analysis also recovered several significant clade-specific effects that are worth discussing. Precipitation was a significant factor shared between amphibians and ray-finned fishes, with polyploid species more common in drier environments. This trend is most evident in amphibians, where polyploid species consistently occur in drier conditions than diploids within their genus, including Neobatrachus in Australia, Phyllomedusa in South America, Xenopus in Africa, Dryophytes in North America, and Pelophylax throughout Europe and parts of Asia (Fig. 1). It has previously been noted that polyploidy facilitates niche shifts into drier climates in Neobatrachus, and coupled with the result here suggests that polyploidy is a common mechanism for arid climate adaptation in amphibians (61). Actinopterygian polyploids were similarly linked to areas of reduced precipitation. Precipitation is known to covary with latitude, so it is possible this observed trend is simply a correlation artifact. However, there are specific arid regions where both amphibian and actinopterygian polyploids appear with disproportionate frequency. In particular, xeric shrublands, grasslands, and steppes throughout Western, Central, and Eastern Asia exhibit some of the highest polyploid frequencies. Of the seven aquatic ecoregions with majority actinopterygian polyploid species two occur in Iran and one encompasses Western Mongolia. In Amphibia, ecoregions such as the Persian Gulf desert, East Afghan montane conifer forests, and three separate ecoregions stretching across the northern Himalayan foothills have only polyploid species. Unfortunately sampling in this area is extremely sparse, often with just one species identified per ecoregion, limiting statistical power. In a related patten, ray-finned fish polyploids are also significantly associated with high-altitude environments. The now polyphyletic schizothoracine fishes (known as “mountain carps”) are known polyploids (62) distributed throughout the Tibetan plateau, and show molecular signatures of high altitude adaptation (63). Additionally, the majority-polyploid Salmonidae are common throughout mountainous regions in northwest North America. As with aridity in amphibians, the repeated association of high-altitude environments and polyploidy in Actinopterygii across clades and continents may indicate a common adaptive pathway. Finally, species richness is often cited as having a negative relationship with polyploid incidence due to competition from diploids and cross-ploidy mating. I found limited evidence to support this idea, as species richness was not significant between amphibian diploids and polyploids. There was a significant negative association in Actinopterygii, albeit with relatively low Cohen’s d. However, large and significant differences were found in Insecta (Fig. 3). One possible element that may help explain this discrepancy is the relationship between parthenogenesis and polyploidy. Polyploidy is tightly linked with thelytokous parthenogenesis in several animal clades including insects but not amphibians or ray-finned fishes (1). Parthenogenic organisms are also predicted to inhabit species-poor areas, for similar reasons as polyploids (64).

Comparisons between polyploid and diploid species revealed several large, significant differences between environmental variables across clades. To see on which of these variables ploidy had the largest impact, I performed a relative importance analysis. Briefly, a series of multiplicative phylogenetic multivariate analysis of variance (MANOVA) models (65) were constructed from every combination of each significant variable for each clade. This resulted in 127 models for Amphibia, 4,095 models for Actinopterygii, and 255 models for Insecta. Within each clade the Akaike information criterion (AIC) for each model was then compared to explore which variables contribute most to the differences between polyploids and diploids. Models that only included glaciation were the best fit in Amphibia and Insecta, whereas a multiplicative model including glaciation and land use was the best fit for the Actinopterygii dataset (Table 1). Each of these models far outperformed any other, with Akaike weights (66) approaching one. The glaciation variable also has the largest relative importance in each clade (Dataset S1).

Across clades, glaciation best explains the discrepancy between the polypoid and diploid environment. Polyploids are more likely to be found in areas that were glaciated at the LGM. This finding is supported by the effect size analysis (Fig. 3) and also serves to explain the latitudinal polyploid gradient. This result contributes to a model of evolution under which recently deglaciated areas provide environments that are conducive to polyploidy. The expanded genomic toolkit of polyploids may enable them to more rapidly adapt to new environments, as has been demonstrated recently in frogs (61) and switchgrass (67). Migrating into new environments also allows polyploids to “escape” the range of their diploid counterparts, allowing them to avoid cross-ploidy competition and mating which is predicted to favor diploids. While deglaciation is the largest and most obvious mechanism through which species are exposed to new environments, human introduction is another example. Invasiveness is commonly associated with plants, and polyploid plants are 20% more likely to be invasive than diploids (68). I recovered this finding in part; polyploids had more invasive species than diploids in all three clades though chi-squared tests were not significant (P > 0.05) in insects, perhaps due to the low sample size of polyploid invasives (one species) (Dataset S2). Anthropogenic transformations such as land clearance and fertilizer application might similarly represent “new” environments more favorable to polyploids (26, 35). In Actinopterygii, polyploids were more common than diploids in areas used for agriculture, and the agricultural land use variable appeared alongside glaciation in the highest-ranking model (Table 1). However, more work is required to disentangle this effect from potential covariates.

The Global Biodiversity Information Facility (GBIF) database was queried for occurrence data on May 23, 2022. Three separate searches were performed, with each restricted to “presence only” data occurrences from one of the three clades under consideration by this study (Amphibia, Actinopterygii, and Insecta). Occurrences with the GBIF ‘basis of record’ field living specimen were excluded, as these often describe animals from zoos or aquariums. Fossil specimens were similarly excluded. Occurrences flagged internally by GBIF for geospatial errors were also excluded from the search. Each dataset is publicly available with the following DOIs (Amphibia: https://doi.org/10.15468/dl.2jxpma, Actinopterygii:https://doi.org/10.15468/dl.6vtqy8, and Insecta:https://doi.org/10.15468/dl.6xx873). After each dataset was downloaded, further filtering steps were performed. Coordinates with a reported uncertainty greater than 100 km were removed, as well as those with a precision of less than one decimal place. Next, the R package CoordinateCleaner (69) (v2.0.2) was employed to remove doubtful occurrences, such as those with equal latitude and longitude or which occur in country capitals, centroids, biodiversity institutions, or GBIF headquarters. Species names were reconciled with the Open Tree of Life reference taxonomy (70) (v3.3). Last, to avoid overrepresentation from specific sampling sites or sources, all coordinates within each species were rounded to one decimal place and duplicate entries were removed. The final filtered dataset encompasses 32,880,023 occurrences.

All major analyses reported in this study were performed using phylogenetic comparative methods. Phylogenetic trees were downloaded from previously published studies for each of the three focal clades. For Amphibia, the tree of Jetz and Pyron (71) was selected, the reference amphibian phylogeny used by the VertLife project. The Actinopterygii tree was taken from Rabosky et al. (72) and is also programmatically available through the R package fishtree. In Insecta, the tree of Chesters (73) was used, a species level phylogeny generated through a stepwise, hierarchical approach using multiple supermatrices. To ensure species labels were analogous with the GBIF occurrence data tip labels for each tree were also reconciled with Open Tree of Life reference taxonomy (70). If species became synonymized after reconciliation, one was retained pseudorandomly, with precedence given to those that shared a genus with their sisters, and the rest were pruned from the tree. These reconciled trees are available at https://doi.org/10.6084/m9.figshare.20457045.

Ploidy estimates were derived from literature, primarily from review articles or published databases (1, 2, 62, 74, 75) (Dataset S3). When two sources disagreed, the most recent was used. Species labeled “polyploid” for the purposes of this study are those in which polyploidy is known to commonly occur in natural populations, though not necessarily exclusively. In two special cases within the amphibian genera Ambystoma and Pelophylax, polyploid kleptons are maintained through hybridization. Due to the phenotypic similarity and range overlap within these complexes, as well as the challenges hybrids pose to standard phylogenetic methods, parent species are considered polyploid as well. I acknowledge this working definition likely includes many diploid occurrences; however, it accomplishes the primary goal of this study by encapsulating areas and environments where polyploid populations are established and maintained. This definition does not include species where polyploids are known only from experimental populations. A few species where polyploidy is considered rare or spontaneous, or whose polyploid status is otherwise uncertain, were excluded from the study. Species not known to be polyploid are presumed diploid, with the exception of a few genera and families where the majority of evaluated taxa are polyploid (SI Appendix, Table S1).

As this study is primarily concerned with differences between diploid and polyploid species, accurate ploidy assignment is paramount. However, such a task is challenging when only a minority of species have had their cytotypes investigated thoroughly. Even with complete information, the lines between diploid and polyploid can be blurred. In order to test how sensitive the results of this study are to the particular methodology and justification used, primary analyses were repeated on two alternative ploidy assignments. First, a stricter definition of polyploidy was implemented, only classifying species as “polyploid” if polyploidy was known to occur in all tested populations exclusively (n = 422), excluding species complexes of variable ploidy or species for whom polyploidy is only common (SI Appendix, Fig. S2A). Secondly, it is probable that several “diploid” species included in this study are in fact polyploids that have not yet been identified as such. To account for this, I included only “diploid” species with records in the Animal Chromosome Count Database (ACC) (76) (n = 3,501) (SI Appendix, Fig. S2B). Under these alternative datasets, the general trends reported by this study were largely recovered although with reduced statistical significance in some cases, possibly as a result of the restricted sample size. One exception to this was in Amphibia, where polyploids occurred at significantly lower latitudes than diploids when the ACC filter was applied. One possible reason for this is the overrepresentation of Xenopus. Xenopus is a polyploid genus (with the exception of one species) of frogs native to sub-Saharan Africa. Due to its relevance to biomedical research, the karyotypes of this genus have been sampled extensively. Whereas Xenopus species represent 0.44% of tropical amphibians in the primary dataset, they represent 11% in the ACC dataset. When Xenopus species are removed from the ACC dataset a significant (P < 1E−7) polyploid latitudinal gradient reasserts itself. Nevertheless, it is clear more extensive cytotype sampling would significantly improve future investigations into animal polyploid distributions.

Global rasters of environmental variables were drawn from a variety of publicly available sources. It has often been suggested that polyploids have greater ecological tolerances to extreme or challenging environmental conditions, such as the colder, drier climes associated with high latitudes. To explore this idea, I collected data on five climate and environmental variables (mean annual temperature, temperature seasonality, annual precipitation, precipitation seasonality, and altitude) from the WorldClim 2.1 (77) database at 5-min resolution. Temperature variables are of particular interest, as extreme temperatures may be associated with a greater incidence of polyploidy, not due to broader ecological tolerances, but because cold or heat shocks simply create more polyploid gametes. To further test this idea, I also calculated how climate variables have changed since the LGM, as estimated by the Community Climate System Model 4 (78), at the same resolution. Similarly, I also collected data on estimated glacial cover at the LGM (79). Glacier shape files were rasterized, with glaciated cells receiving a value of 1, unglaciated cells receiving a value of 0, and cells located in the ocean receiving a value of “not available” (NA). It has been suggested that polyploids are associated with new environments not due to their increased adaptive potential but due to the absence of other species, as interspecific competition and cross-ploidy matings are predicted to adversely impact polyploid populations (2, 33, 56, 57). For estimates of species richness, I summed global rasters for amphibians, mammals, and birds species provided by BiodiversityMapping.org (80). A limitation of this approach is the absence of any nonvertebrate clades, whose patterns of global richness may differ from vertebrates (81). However, I am not aware of any global estimates of species richness in nonvertebrates that would be appropriate for this study at the time of writing. I also fail to include any marine clades, as including estimates of exclusively terrestrial or marine clades together within the same raster would introduce considerable bias. As all polyploid ray-finned fishes identified by this study are freshwater, I elected to use the same terrestrial richness data for all three focal clades to allow comparisons across them. While this dataset may not completely reflect true biodiversity, it does capture the major global patterns (such as the latitudinal diversity gradient) of interest to this study. Finally, polyploids’ expected ecological tolerance has been predicted to extend to areas that have been transformed by human activities as well, with polyploids more resilient to anthropogenic impacts (26, 34, 35). For this hypothesis, I utilized the human footprint index provided by Sanderson et al. (82), as well as agricultural land use datasets from NASA’s Socioeconomic Data and Applications Center (83). All occurrence coordinates and rasters presented and discussed in this study use the WGS 84 reference system; however, visualizations are provided as Robinson projections.

Values from each of the 14 environmental variables were paired with each species occurrence using the R package raster (84) (v3.5.9) and averaged across species. To get relative measures of the environmental differences between polyploids and diploids that are comparable across variables and clades, Cohen’s d was calculated for all relevant comparisons using the R package effsize (85) (v0.8.1). Differences took the form: (x_polyploid – x_diploid), such that positive values indicate that the variable is higher for polyploids than for diploids, and negative values represent the inverse. A species’ ploidy level and geographic range are expected to carry strong phylogenetic signal and, as a result efforts must be taken to distinguish meaningful relationships from covariance due to shared evolutionary history (36, 86). To achieve this, phylogenetic ANOVAs were performed (65), under which ANOVA test statistics are compared to a null distribution simulated from a Brownian-motion model. Phylogenetic ANOVAs were performed for each variable using the R package geiger (87) (v2.0.9), and tested against a null distribution generated from 1,000 phylogenetic simulations. As polyploidy is rare in animals, there is a concern that unbalanced sample sizes violate the homogenous variance assumption of ANOVA tests. To ensure tests were still able to distinguish between hypotheses, additional simulations were performed under which polyploid labels were assigned randomly while maintaining the same sample size. Under this null dataset, no variable was found significant in any clade. Similarly, since the ranges for the majority of diploid species do not overlap with any deglaciated areas, the glaciation dataset is heavily zero-weighted and so violates the normality assumption. To address this, I also performed chi-squared tests between ploidy and glaciation where each species was labelled ‘glaciated’ (≥50% of occurrences within deglaciated area) or ‘not glaciated’ (<50% of occurrences within deglaciated area) which were significant (P < 0.05) in each clade. After ANOVA testing, relative importance analysis was performed. For each clade, a series of phylogenetic multiplicative MANOVA models were created. A model was constructed for every combination of variables that were found significant under the ANOVA tests. AIC values were then compared within each clade to determine the best fitting models.

K.T.D. designed research; performed research; analyzed data; and wrote the paper.

The author declares no competing interest.