Eco-genomic analysis of the poleward range expansion of the wasp spider Argiope bruennichi shows rapid adaptation and genomic admixture
Abstract
Poleward range expansions are commonly attributed to global change, but could alternatively be driven by rapid evolutionary adaptation. A well-documented example of a range expansion during the past decades is provided by the European wasp spider Argiope bruennichi. Using ecological niche modeling, thermal tolerance experiments and a genome-wide analysis of gene expression divergence, we show that invasive populations have adapted to novel climatic conditions in the course of their expansion. Their climatic niche shift is mirrored in an increased cold tolerance and a population-specific and functionally differentiated gene expression response. We generated an Argiope reference genome sequence and used population genome resequencing to assess genomic changes associated with the new climatic adaptations. We find clear genetic differentiation and a significant admixture with alleles from East Asian populations in the invasive Northern European populations. Population genetic modeling suggests that at least some of these introgressing alleles have contributed to the new adaptations during the expansion. Our results thus confirm the notion that range expansions are not a simple consequence of climate change, but are accompanied by fast genetic changes and adaptations that may be fuelled through admixture between long separated lineages.
Introduction
A loss of biodiversity is one of the most commonly forecasted effects of climate change on ecosystems. Confronted with increasingly unsuitable conditions, countless species are believed to face extinction in the coming decades (Thomas et al., 2004; Parmesan, 2006). Recent studies, however, indicate that some taxa have the capacity to cope with environmental change through rapid evolution (Parmesan, 2006; Hoffmann & Sgrò, 2011). Adaptation might even allow a species to emerge as profiteer from global change by enabling rapid range expansions into formerly unsuitable areas (Clements & Ditommaso, 2011; Hill et al., 2011; Franks & Hoffmann, 2012). Such contemporary adaptation has been well-studied on the phenotypic level and several examples have been brought forward, which often involve the evolution of increased dispersal or phenological shifts (Huey et al., 2000; Hill et al., 2011; Parmesan, 2006; Thomas et al., 2001; Phillips et al., 2006; Bradshaw & Holzapfel, 2001). Another key adaptation concerns tolerance toward novel climatic conditions (Colautti & Barrett, 2013; Jeong et al., 2014), for example colder temperatures at the former range edge (Krehenwinkel & Tautz, 2013). While examples for rapid adaptation during range expansions are accumulating, its functional genetic background remains to be elucidated. To date only few examples from model organisms provide insights into the molecular background of such rapid adaptation along environmental gradients (e.g. Hoffmann & Willi, 2008; Schmidt et al., 2008).
In small and scattered range edge populations, mutational input or standing variation has been considered to be too low to allow fast adaptation to a rapidly changing environment (Bridle & Vines, 2007; Pujol & Pannell, 2008). In such cases, admixture of formerly isolated genetic lineages could provide new genetic variation and facilitate evolutionary change (Mallet, 2007; Nolte & Tautz, 2010; Abbott et al., 2013). Such hybridization may be of particular importance during colonization of new habitats, where the advantages of novel variation could outweigh the effects of genetic incompatibilities (Verhoeven et al., 2011; Rius & Darling, 2014). Recent years have seen an accumulation of studies associating genetic admixture and successful range expansions on invasions (e.g. Nolte et al., 2005; De Carvalho et al., 2010; Keller & Taylor, 2010; Lucek et al., 2010). However, a comprehensive analysis linking ecological divergence and genome-wide genetic variation is still lacking.
The European wasp spider Argiope bruennichi constitutes a striking case of a massive recent range expansion. As a large, conspicuously colored and easily identifiable spider species, its spread has been exceptionally well-documented over the past century (Guttmann, 1979; Kumschick et al., 2011). From around 1930 onwards, the wasp spider started expanding its range into increasingly Continental climate regions (Guttmann, 1979; Kumschick et al., 2011). Originally largely distributed in Mediterranean and Oceanic regions in Southern Europe, it can now be found as far north as Finland. Based on analyses of historical museum specimens, we have previously shown that the onset of the invasion is tightly correlated with an increase of genetic and morphological variation in expanding populations. Further analyses suggest that the expanding lineages may have emerged out of an initial hybrid swarm of European and Asian wasp spider populations, which might have adapted to novel climatic conditions at the species’ former range edge (Krehenwinkel & Tautz, 2013). However, the actual ecological differentiation of native and invasive wasp spider populations, the genome-wide extent of admixture and the molecular signatures of adaptation during their expansion remain to be explored.
For this reason, we have generated an ecological niche model for expansive and native European spider populations and show that they indeed occupy significantly different climatic niches, with winter temperature as a major differentiating factor. To test this model prediction, we performed a cold tolerance experiment, mimicking an extreme winter night that may be most relevant for survival in Northern latitudes. We found that there are indeed differential cold adaptations in the respective populations. Using transcriptomic analysis, we show that these are also reflected in genome-wide, population specific differential gene expression patterns. Finally, we use whole genome sequencing of several populations of the wasp spider to quantify the admixture during the range expansion and assess to what extent this might have been adaptive. We find clear genomic divergence between the ecologically divergent invasive Northern and native Southern European spider populations, supporting a pattern of genome-wide environmental isolation. The genetic differentiation is associated with an introgression of East Asian alleles, some of which may have contributed to the new adaptations.
Material and methods
Species distribution models
We performed ecological niche modeling for native and invasive populations based on a comprehensive set of 203 native species records and 134 invasive species records, which were compiled through own field work and from the online data base Global Biodiversity Information Facility, wherein only records matching the temporal resolution of the climate data were used for further processing (see below). Invasive populations comprise those that have been established in the course of the species’ range expansion. Genetic and phenotypic analyses have been used to define the border between native and expansive populations (Krehenwinkel & Tautz, 2013). To remove possible negative effects of spatial autocorrelation as measured by Moran's I, a multivariate semi-variogramm based on the set of climate predictors was computed to determine the minimum radial distance between species records. After spatial subsampling, 113 native and 49 invasive records remained for further analyses.
As climate predictors we used a comprehensive set of 19 bioclimatic variables (Table S1) with a spatial resolution of 2.5 arc min (about 4 km in the study area) available through http://www.worldclim.org (Hijmans et al., 2005). As potential evapotranspiration may be an important climate factor affecting the physiology of the spiders, we computed equivalents to bio1, bio4–7, bio10, and bio11 based on monthly variables provided by Trabucco & Zomer (2009). Based on the total set of 26 variables (19 worldclim + 7 potential evapotranspiration) we computed pairwise spearman rank correlations to assess multicolinearity, wherein species distribution models (SDMs) were developed with a subset of variables with R2 < 0.75: bio2, bio7, bio8, bio10, bio12, bio14, bio15, bio18, bio19, and PET_HE_bio7.
As modeling framework we used the biomod2 platform for Cran r (Thuiller et al., 2014). A weighted ensemble model was computed based on repetitive runs of six algorithms (Generalized Linear Models, Generalized Boosting Models, Generalized Additive Model, Multiple Adaptive Regression Splines, Artificial Neural Networks and Maxent). Therefore, three different sets of each 1000 pseudo-absences were generated using the ‘SRE’ option in biomod2, for each of these sets the species records were three times split in 80% used for model training and 20%, which were used for evaluation of model performance in terms of the Area Under the receiver operating characteristic Curve (AUC, Swets, 1988), Cohen's Kappa and the True Skill Statistic (Allouche et al., 2006). The final weighted ensemble was computed based on all single models with AUC > 0.7, wherein the better performing models were considered to a higher degree in the final ensemble. The decay factor was set to 1.6. Relative importance of predictors was assessed internally in biomod2 using a permutation approach.
Differentiation between native and invasive populations of the spider in realized climate niche space were assessed twofold: first using the PCA-env approach of Broennimann et al. (2012) and second using the multidimensional extension of this technique as recently proposed by Blonder (2014b).
Available climate space in the PCA-env approach, defined the complete climate space within a radial buffer of 100 km enclosing the respective records as described by the 26 variables. Niche overlap in terms of Schoener's D is computed based on kernel density estimates for both native and invasive species records in PCA space accounting at the same time for the relative frequency of specific climate conditions within the study area. Randomization tests for niche equivalency and similarity as proposed by Warren et al. (2008) were computed to test the hypothesis of niche equivalency (niches occupied by native and invasive populations are identical) and niche similarity (niches occupied by native and invasive populations are more similar than can be expected by chance given the available niche spaces for both groups).
Following Blonder et al. (2014a), we computed the n-dimensional hypervolumes occupied by both native and invasive populations based on PCA space spanned by four PCs with eigenvalues >1, applying a bandwidth of 0.5. This technique extends the framework proposed by Broennimann et al. (2012) by applying multidimensional kernel density estimators to derive a density distribution of species records in PCA space, which is used to compute the total volume of the niche space of native and invasive populations as well as the intersection of both volumes. The results allow a quantification of those parts of niche space occupied by both groups as well as the unique parts of native and invasive populations. Niche overlap was estimated by the Sørensen index based on unique and shared hypervolumes, wherein all computations were performed in Cran r using the relevant functions of the hypervolume package (Blonder, 2014b).
Cold tolerance experiment
Mated wasp spider females were collected in August 2012 in native European populations in Southern Portugal, and recently established invasive ones from Sweden, Latvia, and Estonia. The spiders were kept in the laboratory at room temperature (20 °C) in 200 mL plastic cups, sprayed with water every second day and supplied with house flies, until they constructed an egg sac. We initially set up a day and night cycle of 14 h light followed by 10 h of darkness. All egg sacs were then kept at room temperature (20 °C) for 4 weeks and sprayed with water every day. Subsequently, the temperature was lowered to 15 °C, the light cycle reverted to 10 h of light and the egg sacs kept for additional 2 weeks. We then opened the eggsacs and forced the spiderlings to emerge. Usually, wasp spider offspring will remain in a diapause state in the protective silk case for the whole winter. However, to test cold tolerance in early instars at the beginning of the diapause, we had to force their emergence. Each ten siblings per eggsacs were then transferred into a petridish and exposed to −20 °C for 6 h in a common lab freezer. After the long term freezing, the petridishes were transferred to a climate chamber tempered at 22 °C and the spiders allowed to recover from cold shock. We counted the fraction of recovering spiders in 30 min intervals for the next 2 h. Recovering was scored, if the cold shocked spider started moving its legs again. After 2 h, the petridishes were left for another 22 h at 22 °C and the amount of damaged or dead spiderlings per eggsac counted. Damaged spiderlings were those which had recovered, but showed problems walking properly. We thus estimated the overall fraction of survivors after long term freezing for Northern invasive and Southern native European populations, as well as the temporal cold shock recovery rate over 2 h.
Generation of a wasp spider reference genome assembly
In order to provide a suitable reference genome for our study, we had to generate a genome sequence of the target species A. bruennichi. The quality of a de novo assembly can be greatly improved by reducing the amount of heterozygous loci, for example by sequencing DNA of an inbred specimen (Vinson et al., 2005). As we did not have inbred lines, we relied on a wild caught specimen from a low diversity population of the wasp spider on the island of Madeira (see Table S2 for sampling sites). DNA was extracted from the whole specimen using the ArchivePure blood and tissue kit (5 PRIME, Hamburg, Germany). An RNA digestion step was included using RNAse A solution (7000 U mL−1; 5 PRIME). The extraction was carried out according to the manufacturer's protocol. A paired end DNA library with an average insert size of 250 bp was prepared by the Center of Genomics at the University of Cologne. The library was sequenced on one lane of an Illumina HiSeq 2000 to an approximately 20-fold coverage according to the manufacturer's protocol (Illumina, San Diego, CA, USA). As we did not know the exact genome size of A. bruennichi, we relied on available data for its sister species Argiope aurantia and other members of the genus Argiope. All of them have similar genome sizes of about 1.5 Gb (Gregory, 2014).
The raw data was quality trimmed using PoPoolation (Kofler et al., 2011a), using a minimum quality of 20. The trimmed data was then assembled using CLC Genomics Workbench (CLC Bio, Cambridge, MA, USA), with a word size of 45, a bubble size of 98, a minimum contig length of 1000 and including a scaffolding step. Repetitive DNA was masked out from the resulting assembly using RepeatMasker (Smit et al., 1996–2010) under default parameters. Moreover, we used Blastn to identify and remove bacterial contamination in the final genome assembly. Several bacterial genomes (Thermoanaerobacter italicus, Staphylococcus aureus, Escherichia coli, Bacillus cereus) served as a reference at an E-value cutoff of 10−3. We generated a final genome assembly of 240 061 contigs with an N50 size of 11 kb.
Generation of wasp spider reference transcriptome assembly
To recover as many genes as possible from the wasp spider's genome, we generated a reference transcriptome based on an ontogenetic series of RNA samples. We prepared four different total RNA extractions using the Qiagen RNeasy Kit according to the manufacturer's protocol (Qiagen, Hilden, Germany). A DNAse digestion step was carried out using the Qiagen RNase-Free DNase Set. Separate extractions were prepared from a whole adult male, a whole adult female, 20 eggs and 20 spiders in the first nymphal instar. Female spiders are usually well-nourished and their opisthosoma is filled with eggs and digested food remains. We consequently extracted the female in two separate reactions, one containing the prosoma including the legs and one containing the opisthosoma. All extractions were done with specimens from a population from Plön, Schleswig Holstein, Germany, which belongs to the invasive range of the species (see Table S2 for sampling sites). The extraction series thus covered early and late developmental stages, as well as sex specific differences in gene expression. The RNA extracts were sent to the Center for Genomics at the University of Cologne, where cDNA libraries with an average insert size of 250 bp were prepared. These libraries were then sequenced on one lane of an Illumina HiSeq 2000, as described above for the genome data.
Quality trimming was performed as described for the genomic DNA. The paired reads were assembled using SOAP de Novo trans (Xie et al., 2014) with a maximum read length of 100 bp, a k-mer size of 25 and including a mapping and a scaffolding step. Subsequently, we used cd-hit-est (Li & Godzik, 2006) with a similarity cutoff of 90% to remove redundant contigs from the de novo assembly and to generate scaffolds. We then translated the transcriptome using transdecoder (http://transdecoder.sourceforge.net/). The remaining transcriptome dataset was blasted and annotated (using blastx with an E-value cutoff of 10−3) with Blast2Go (Conesa et al., 2005). The final reference RNA assembly was then aligned to the previously generated reference genome using blat (Kent, 2002) under default settings and at a similarity threshold of 98% and a minimum overlap of 100 bp. For further analysis we only included transcripts with a significant blast hit, which map into a genomic contig. This approach allowed assembling a reference transcriptome with an N50 of 2 kb, of which a total of 12 946 transcripts could be mapped to the genome and showed a significant blast hit in the database, that is correspond to genes known from other species. Hence, we exclude spider-specific orphan genes from this analysis at present, since it would currently not be possible to derive functional inferences for them. More details on the transcriptome assembly statistics can be found in Krehenwinkel (2013).
Differential gene expression between distinct thermal regimes in native and expansive wasp spider populations
Each six egg sacs from the Baltic (invasive sample) and Portuguese (native sample) populations used in the cold tolerance experiment (see above, see Table S2 for sampling sites) were selected for the reciprocal transplant RNA sequencing experiment. We chose egg sacs, which had been built only a few days apart, to reduce age effects in the experiment. Each egg sac was opened and the spiderlings forced to emerge. Sixty spiders per egg sac were then evenly distributed to three Petri dishes (20 spiders per Petri dish), equipped with a slightly wetted paper tissue. One dish per egg sac was transferred into a Memmert IPP800 thermal chamber (Memmert, Schwabach, Germany) at 15 °C. Three distinct thermal profiles were set for the incubators. A control treatment was permanently set to 15 °C, a cold treatment started at 15 °C and lowered by 5 °C every 2 h. A heat treatment started at 15 °C and increased by 5 °C every 2 h. The spiders were kept under these conditions for 10 h, allowing the cold treatment to reach −10 °C and the heat treatment to peak at 40 °C. The Petri dishes were then taken out of the thermal chambers. The spiderlings were immediately frozen in liquid nitrogen and stored on dry ice. Total RNA was extracted from the spiders and samples were sent to the Center for Genomics in Cologne for library preparation and sequencing as described above. The 36 samples were sequenced on two lanes of an Illumina HiSeq 2000. RNA read counts were generated using CLC genomics workbench, by aligning reads to the previously generated reference transcriptome with a mismatch cost of 1, insertion cost of 3 and deletion cost of 3. A minimum read count of 10 per group was used as threshold for a transcript to be considered in the analysis. Significant differences in gene expression between Northern invasive and Southern native European populations were evaluated for all three experimental treatments using DESeq (Anders & Huber, 2012) at an false discovery rate (FDR) of 0.1. We then performed a GO-term enrichment analysis for the differentially expressed genes, compared to the reference transcriptome using Blast2Go at an FDR of 0.1. We performed GO-term enrichment for two separate datasets. First, we tested for an enrichment for all differentially expressed transcripts between invasive and native populations at each experimental condition. These datasets thus included overlaps of differentially expressed genes between different temperature treatments (this corresponds to full circles in Fig. 3). Then we tested those genes, which were exclusively differentially expressed between invasive and native European populations in one experimental condition.
Sampling of specimens for genome sequencing
We selected samples from two native (Portugal and Italy) and two invasive (Sweden and Baltic) European populations for genome sequencing (see Table S2 for site details). These native and invasive populations were clearly phenotypically differentiated, showed little ecological niche overlap and are exposed to divergent thermal and precipitation regimes. European spiders were sampled between 2010 and 2012. We sampled five specimens from three populations in each region, resulting in 15 specimens per region and 60 spiders in total. DNA was extracted from single animals as described above for the Madeiran reference specimen. The DNA quantity was measured using a ND-3300 Fluorospectrometer (PEQLAB, Erlangen, Germany) according to the manufacturer's protocol. Equal quantities of DNA for each of the 15 specimens per region were pooled. In addition, we included five Japanese specimens as outgroup into our analysis. To cover a wide range of shared East Asian genetic variation, those specimens were sampled from different locations over the Japanese islands (Table S2). All five specimens were sequenced at 20× coverage in separate libraries on an Illumina HiSeq 2000 and the resulting reads mapped as described above. The five specimens were combined into a single alignment, which was treated as a population pool.
Genomic divergence and admixture during a range expansion
Each DNA pool was sequenced to approximately 20-fold genome coverage per sequenced sample. All paired end reads were quality trimmed as described above and then aligned to the Madeiran reference genome using Bowtie2 under default alignment parameters (Langmead & Salzberg, 2012). PoPoolation and PoPoolation2 (Kofler et al., 2011b) were then used to infer genome-wide nucleotide diversity and genetic differentiation. Allele frequency differences were calculated for each variable position in the genome alignment with a minimum coverage of 10. Moreover, pairwise FST and nucleotide diversity were estimated for sliding windows of 5000 bp. Considering the comparably low N50 of only about 10 kb, this small window size was chosen. The analyses were run with a minimum quality per base of 20, at least 0.25 fold coverage of each window, a minimum count of the minor allele of two for each included population and a coverage of 10–40-fold. We used a sliding window approach, as the identification of single alleles from pooled data can be error prone. The according sliding windows thus represent an average out of several SNPs per window. To test the contribution of Eastern Asian variation to the expansion success of the species, we have used a Japanese outgroup. Asian variation probably does not directly enter the European gene pool from Japan, but rather from Central Asian steppe populations. Consequently, we might exclude private Central Asian alleles from our analysis. Nevertheless, Japanese specimens provide an ideal basis to identify ancestry informative SNPs for the Eastern and Western Palearctic lineages. We identified these lineage diagnostic SNPs using a 95% frequency difference cutoff and a minimum coverage of 10 between Japanese and both native Southern European populations. Using these loci, we could estimate the amount of introgression in invasive populations.
For modeling, we used the method implemented in Stemshorn et al. (2011), to test if our data would fit one of two alternative models. (1) A two parameter model, considering founding allele frequencies and time since the hybridization event, assuming the hybrid population to be affected only by drift and no secondary gene flow after hybridization. (2) A four parameter model, considering founding allele frequencies, migrant allele frequencies, migration rates and time since initial admixture. Here, the hybrid population receives immigrants every generation after a recent founding event. We used one lineage diagnostic SNP per genomic contig, reducing the dataset to 17 197 SNPs. We then randomly sorted the dataset repeatedly and generated 50 datasets with 400 SNPs for each Baltic and Swedish hybrid population, including two European (Portugal & Italy) and one Asian (Japan) parental population. We then performed several test runs, to optimize our input parameters and found to achieve consistent results with 200 steps and 40 chains. Next to the identification of a proper population model of the hybridization event, the analysis also allows for the identification of outlier loci, which do not conform to the neutral expectations of the model. In a next step, we used the averaged model output parameters from all 100 analyses as an input for a novel Perl script (F.J. Sedlazeck, J. Cheng, J. Altmüller, A. von Haeseler and A.W. Nolte, in review) to identify outlier SNPs in our datasets. Using the averaged population parameters from all independent datasets as the most probable parameter, this script outputs a P-value for departure from neutrality for every analyzed locus in all 100 datasets. Using an FDR of 0.1, we then identified all loci, which significantly departed from a neutral model. Based on the outlier loci, we could set a cutoff frequency for introgressed Asian alleles in invasive European populations, which is unlikely to have arisen by neutral processes alone. Applying this threshold, we identified outlier contigs as candidates for a potential adaptive divergence.
FST outlier detection
Selective sweeps involved in new adaptations in Northern invasive populations should lead to high differentiation between invasive and native European populations. To identify such regions, we obtained the FST measure for all 5 kb sliding windows between the two Southern native populations and determined the standard deviation. We then filtered for those windows that showed an FST above the twofold standard deviation in every of the four pairwise comparisons between Southern native and Northern invasive populations. We consider this procedure as conservative and stringent, although it does not guarantee that all outliers are indeed due to selective sweeps. Our previous phylogeographic analysis (Krehenwinkel & Tautz, 2013) had suggested a complex and possibly stepwise history of the emergence of the expanding population, which may have included bottlenecks that could lead to random fixation of some haplotypes. On the other hand, the overall diversity analysis provides no indication for a major bottleneck, that is it is unlikely that neutral fixation has caused many of the FST outliers.
Results
Species distribution models
We have generated a species distribution model as well as two multivariate climate niche overlap analyses for native and expansive spider populations in Europe (Fig. 1). We find very little niche overlap between native and invasive populations (Schoener's D = 0.182) according to the classification proposed by Rödder & Engler (2011). Moreover, we can reject the hypothesis of niche equivalency according to the procedure proposed by Warren et al. (2008) as well as the hypothesis of niche similarity when comparing native populations vs. random records from the available niche space of invasive populations (P < 0.05). Analyses of PCA based 4-dimensional hypervolumes of both native and invasive populations indicate significant differences between the niche spaces occupied by native and invasive populations. The niche volume of native populations is much larger (82.15) compared to invasive populations (29.17), with an intersection of only 5.3 suggesting a very low niche overlap. This implies a strong ecological separation of invasive and native spider populations (Figs 1, S1 and S2 and Table S3), which must have developed within a century.
Cold tolerance experiment
The niche model shows that invasive populations have colonized a habitat which is colder and more humid than the original native range. Minimum annual temperature as predicted by the modeling is significantly lower for the invasive than for the native range (Fig. 2a). We had previously shown in thermal preference tests and a reciprocal transplant experiment that relevant differences exist between the Southern native and the Northern invasive populations (Krehenwinkel & Tautz, 2013). Here, we focused on freezing tolerance and conducted a cold stress experiment, exposing invasive Northern and native Southern spiders to 6 h freezing at −20 °C and compared their recovery rates. This experiment shows a much higher cold tolerance of Northern invasive spider populations, compared to Southern native ones. On average, 79% of the native Southern spiders, but only 37% of the invasive Northern ones were damaged or dead 24 h after the cold shock (U-test, P < 0.001; Fig. 2b). Moreover, invasive Northern spiders recover significantly faster from cold shock than native Southern ones. Thirty minutes after the cold treatment, 12% of the invasive, but only 1% of the native spiders had started to move. After 1 h, these numbers had shifted to 41% vs. 8% and after 2 h to 72% vs. 50% (H-test, P < 0.001; Fig. 2c). Hence, this experiment, in combination with our previous results shows that there are clear physiological differences between the populations that are relevant for the fitness in the new climatic niche.
Gene expression response to thermal stress
A genetically based thermal niche shift of invasive Northern spider populations should be reflected in a population specific gene expression response. To test this hypothesis, we mimicked a reciprocal transplant experiment exposing Northern invasive and Southern native European spiders to a different thermal condition (−10 °C = cold stress, 15 °C = normal, 40 °C = heat stress). We then screened for genome-wide differential gene expression based on RNASeq data. We find that both populations show a major transcriptomic response toward increased temperatures vs. very little general change at reduced temperatures, which may simply reflect a general physiological response to heat vs. cold (Fig. 3a). However, the populations also show generally distinct expression differences. In total, we find that 1012 transcripts were differentially expressed in the comparison between native and invasive populations during any of the experimental conditions (Fig. 3b). Among the differentially expressed transcripts, we find a bias toward down-regulated transcripts under normal conditions (250 of 395) and cold stress conditions (375 of 609), but not under the heat treatment (289 of 549). This difference between normal and cold vs., heat conditions is significant (Fisher's exact test, two tailed P < 0.001). Differential expression appears to be condition specific, more than half (630) of the differentially expressed transcript are exclusively identified in one of the experimental conditions. Moreover, we find a strong bias for differential expression during thermal stress. Only 90 transcripts are exclusively differentially expressed during normal conditions, compared to 252 during cold stress and 287 during heat stress. The difference of uniquely differentially expressed transcripts between normal and thermal stress conditions is highly significant (Fisher's exact test, two tailed P < 0.001).
To assess whether any particular functional categories are overrepresented among the differentially expressed genes, we performed a GO-term enrichment analysis. Similar to the gene expression, the GO-term enrichment is condition specific and particularly pronounced during cold stress (Table 1). Most functional enrichment is found for differential expression at cold stress (−10 °C; GO terms: acetoacetate-CoA ligase activity, white fat cell differentiation, response to oleic acid). Moreover, the set of 252 transcripts, which are exclusively differentially expressed during cold stress shows a general down-regulation of ribosomal protein transcripts (Table 1). One GO-category is shared between cold and control conditions (positive regulation of insulin secretion). We do not find any enrichment at heat stress. In conjunction with the large number of heat responsive transcripts within each population, this suggests a rather unspecific response to heat stress. On the other hand, the pronounced functional enrichment for the cold stress indicates an important role for being able to respond to cold stress in the transcriptional divergence of invasive and native populations.
| GO ID | GO name | Category | Treatment | #/# | P |
|---|---|---|---|---|---|
| GO:0032024 | Positive regulation of insulin secretion | Process | −10 °C | 5/9 | 0.02 |
| GO:0032024 | Positive regulation of insulin secretion | Process | 15 °C | 4/9 | 0.07 |
| GO:0030729 | Acetoacetate-CoA ligase activity | Function | −10 °C | 4/7 | 0.053 |
| GO:0050872 | White fat cell differentiation | Process | −10 °C | 4/7 | 0.053 |
| GO:0034201 | Response to oleic acid | Process | −10 °C | 4/8 | 0.094 |
| GO:0003735 | Structural constituent of ribosome | Function | −10 °C unique | 10/83 | 0.0057 |
| GO:0005840 | Ribosome | Component | −10 °C unique | 11/108 | 0.0057 |
| GO:0006412 | Translation | Process | −10 °C unique | 11/144 | 0.047 |
- #/# refers to the number of enriched transcripts in the category of the total number in the category.
Genome-wide genetic differentiation
Our comparative genomic analysis is based on two native Southern (Portugal and Italy) and two recently established invasive Northern populations (Sweden and Baltic). We analyzed 183 008 nonoverlapping sliding windows of 5 kb across the contigs of the reference genome sequence.
The two native Southern and the two invasive Northern populations are both more similar to each other (FST = 0.09 and 0.08) than between them (FST = 0.11) indicating a significant (P < 0.0001, H-test) genetic differentiation. This suggests that they can be treated as closely related, yet differentiated genetic lineages. The larger similarity among the native Southern populations is likely due to ongoing gene flow, while the similarity among the invasive Northern populations could be due to their recent emergence with a joint history. The intrapopulational similarities vs. the interpopulational differences become particularly clear when assessing the FST-value distribution across all 5 kb windows (Fig. 4a). The native Southern and invasive Northern populations show a very similar distribution among each other, but distinctive differences between them. Intriguingly, however, although the invasive Northern populations are the expansive ones, they show a significantly increased genetic diversity compared to the native Southern ones (πNorth = 0.007, πSouth = 0.005; U-test, P < 0.001), which is also reflected in the distribution of nucleotide diversity classes in the 5 kb window analysis (Fig. 4b).
We had previously noted this pattern of increased diversity in the expanding lineages based on mitochondrial variants and a small set of SNPs and were able to show that this might be due to introgression of East Asian genetic material (Krehenwinkel & Tautz, 2013). We therefore generated resequencing genome data for populations from Japan and used these for comparisons with the European populations. The genome wide analysis of FST divergence shows that Japanese populations are indeed very distinct from European ones. However, invasive Northern European spider populations are significantly less divergent from Japanese ones, than their native Southern European relatives (FST = 0.4 vs. 0.45, P < 0.001, U-test). When assessing the frequency of lineage diagnostic SNPs (defined as a SNP frequency difference > 0.95) we find many more for the comparison of Japan with the native Southern populations (17 293) than the invasive Northern populations (3151; Fig. 5a). One can thus take the diagnostic SNPs distinguishing the native Southern populations from the Japanese population to ask at which frequency they occur in the invasive Northern populations as a proxy for the degree of influx of Asian genetic material. Figure 5b shows the corresponding distribution with the majority of these SNPs occurring at a low frequency and only few close to fixation.
Population genetic modeling and FST outliers
To be able to assess whether the frequency distribution of introgressing East Asian lineage diagnostic SNPs into the invasive Northern European populations is compatible with a neutral immigration scenario or whether adaptive introgression has to be invoked, we employed a population genetic modeling approach. This approach was originally developed in the context of studying a recent admixture of fish species in the Rhine system (Stemshorn et al., 2011). It is based on modeling the frequency distribution of diagnostic SNPs under two different scenarios. The first scenario assumes a single hybridization event at some point in time without further gene flow, while the second assumes continued gene flow after the initial hybridization. Based on extensive simulation runs, we find the second model to better explain the allele frequency distributions (log likelihood values (ln(PP(D|x))model 2 = 494.5 ± 48.5, ln(PP(D|x))model 1 = 366.8 ± 27.2), P < 0.001, U-test). However, since the second model includes more parameters, a better fit could simply be due to the higher parameterization. Nonetheless, one can use the identified model parameters (Fig. S3) to estimate the proportion of SNPs that fall outside a simulated purely neutral model of admixture. At an FDR of 0.1, the neutral simulation would suggest that an introgressed lineage diagnostic allele would not be expected to reach a frequency of more than 50%, that is any SNP alleles with higher frequency are candidates for an adaptive introgression. In our data, this applies to about 0.5% of the introgressing lineage diagnostic SNPs, which map into 1003 genomic contigs.
Genomic regions that are involved in recent adaptations can also be identified by selective sweep signatures that raise their FST. To detect such regions, we searched for all genomic contigs with at least one 5 kb window showing a consistently high FST between the native Southern and invasive Northern populations. We find a total of 962 such FST outlier contigs, 264 of which overlap with the 1003 contigs identified as introgression outliers. At the same time, these FST outliers are mostly located in regions of very low genetic differentiation within invasive and within native European populations.
The number of genes mapping into both outlier fractions is significantly increased compared to the whole genome (7% of the genomic contigs map a transcript, 39% of the introgressing outliers and 44% of the FST outliers; Chi square test, χ2 = 2208.29, two-tailed P < 0.0001) and is further significantly increased in the overlap fraction (55% contigs with a transcript; Chi square test, χ2 = 15.964, two-tailed P < 0.0001). The GO-term enrichment analysis shows a significant functional enrichment of GO terms mostly for the FST outliers (Table 2).
| Outlier-category | GO-ID | Term | GO-category | #/# | P |
|---|---|---|---|---|---|
| FST outlier | GO:0005545 | 1-phosphatidylinositol binding | Function | 5/7 | 0.0054 |
| FST outlier | GO:0046498 | S-adenosylhomocysteine metabolic process | Process | 4/5 | 0.017 |
| FST outliert | GO:0048268 | Clathrin coat assembly | Process | 5/5 | 0.017 |
| FST outlier | GO:0030276 | Clathrin binding | Function | 5/6 | 0.017 |
| FST outlier | GO:0004719 | Protein-l-isoaspartate (d-aspartate) O-methyltransferase activity | Function | 4/6 | 0.017 |
| FST outlier | GO:0004357 | Glutamate-cysteine ligase activity | Function | 3/3 | 0.021 |
| FST outlier | GO:0034334 | Adherens junction maintenance | Process | 3/3 | 0.021 |
| FST outlier | GO:0000096 | Sulfur amino acid metabolic process | Process | 4/4 | 0.031 |
| FST outlier | GO:0046500 | S-adenosylmethionine metabolic process | Process | 4/4 | 0.031 |
| FST outlier | GO:0006750 | Glutathione biosynthetic process | Process | 3/4 | 0.049 |
| FST outlier | GO:0005088 | Ras guanyl-nucleotide exchange factor activity | Function | 5/17 | 0.05 |
| Introgr. outlier | GO:0048149 | Behavioral response to ethanol | Process | 4/6 | 0.074 |
- #/# refers to the number of enriched transcripts in the category out of the total number in the category.
Discussion
The present study had two aims: (1) to assess climatic niche differences relevant for new adaptations during the expansion of the spider populations and (2) to trace genomic and transcriptomic changes associated with these adaptations. The climatic modeling shows that the invasive Northern populations occupy a distinct new climate zone that specifically required the development of a better cold adaptation. Our laboratory experiments on freezing tolerance show that this is indeed the case. Animals from invasive Northern populations show a much better recovery and survival rate after exposure to extreme conditions, as they would occur in winter times. Furthermore, they show distinct sets of differentially expressed genes in response to different temperature treatments. With respect to cold adaptations, it is of particular interest to observe that the ribosomal machinery appears to be specifically down-regulated upon cold treatment in the invasive Northern populations. The regulation of ribosomal expression has been associated with the diapause auf Aedes aegypti mosquitoes (Robich et al., 2007) and could thus also be a central part of the wasp spider's diapause. Northern European populations might be equipped with a modified diapause response that could be an important adaptation to cope with colder winter climate. A general reduction of metabolic activity is a central element of diapause (Poelchau et al., 2011). By reducing the number of ribosomes, the invasive Northern spiders might be able to quickly adjust protein production and slow down their metabolic activity. Such a response could be regulated by a single regulatory trigger, that is has the potential to evolve very fast. However, we also observe a number of other population-specific changes in gene expression that do not point directly to underlying mechanisms, but suggest a polygenic adaptive response.
The full genome analysis shows a clear genetic divergence of invasive and native spider populations, indicating a rapid isolation of populations by environment (Sexton et al., 2014). The genome analysis also confirmed the genomic introgression pattern that we had previously proposed. We find that the invasive Northern populations have taken up a considerable amount of genetic material from Eastern Asian populations. This raises the possibility that admixture has driven the novel adaptations, but our population genetic modeling suggests that more than 99% of the introgressed alleles are compatible with a neutral introgression scenario. Hence, only a small fraction of outliers are detected that may have introgressed adaptively, but this involves still several hundred genes. Furthermore, we need to emphasize that our analysis was restricted to the subset of SNPs that are lineage diagnostic between the unrelated native populations in Southern Europe and Japan. We do not expect that the Japanese populations were the direct source of the introgressing alleles. Instead, the actual source of the Asian variation in invasive Northern Europe are most likely Central Asian populations. Interestingly, the continental Central Asian steppe climate is also characterized by cold winters. But since there are currently only few samples from these populations, a more detailed population genetic modeling cannot be done at present. Hence, it seems possible that preadapted alleles or haplotypes that confer cold resistance have come from these populations and are present at higher frequencies in the invasive Northern populations than we can currently infer from the Japanese lineage diagnostic SNP distribution. Many of the private alleles in Northern European populations, including the observed FST outliers, might thus constitute variation from Central Asia. Only a detailed sampling over the Central Asian range of the spider and a corresponding genome analysis could confirm this.
The complex population history also limits our FST outlier analysis approach. Although true selective sweeps indicative of a recent adaptation should be among this outlier fraction, drift during bottlenecks can also create outlier signals. Population genetic modeling based on site frequency spectrum analysis (Excoffier et al., 2013) could potentially resolve this question. But this requires sequences from individuals and denser population sampling with respect to Asian populations to allow reliable estimates. On the other hand, the fact that transcripts and GO terms are enriched in the FST outlier fraction, suggests that it is not simply composed of randomly fixed regions. Hence, while the unequivocal detection of genes that have contributed to the adaptations in the invasive Northern populations will require much more work, our analysis suggests that an introduction of novel genetic variation is a major driver of their divergence (Abbott et al., 2013). Still, we cannot exclude that a utilization of standing variation (Barrett & Schluter, 2008) has played a role as well. Considering the high dispersal ability of wasp spiders, a maintenance of genetic variation during a range expansion, without a pronounced colonization bottleneck is actually very likely. Young spiders are capable of travelling vast distances by wind mediated passive transport on silk threads (‘ballooning’), and can build up large founding populations (Foelix, 2010). Thus, they probably maintain a considerable fraction of the native genetic variation, contributing to the high diversity observed in invasive populations.
But independent of the exact history of the origin of the alleles that confer the cold adaptation, our results allow two important conclusions with respect to understanding the patterns of population expansions. The first is that although global change may serve as a trigger, it may not be the sole cause for the pole-ward expansion of populations. Rather, the expansion is associated with new adaptations that evolve very fast. Second, the genetic variation required for the new adaptations does not necessarily have to be derived from the standing variation only, but could be supplemented through the mixture of genetic lineages that had long been separated. This latter finding is particularly relevant for the discussion around the role of the genetic diversity in rear edge populations, that is those that stay behind after an expansion. It has been suggested that such populations might harbor high levels of genetic variation, and that loss of these populations could cause a disproportionately large reduction in overall genetic diversity in such taxa (Hampe & Petit, 2005). We find that admixture from previously separated populations can replenish variation such that the expanding population actually has a higher diversity than its original source population.
It has long been assumed that such admixture might prevent the establishment of novel adaptations (Bridle & Vines, 2007) or could even contribute to extinction (Rhymer & Simberloff, 1996). But in the case of wasp spiders, it is evident that the admixture has not hampered the expansion and may have even fuelled it. This places A. bruennichi in line with a growing number of examples of the interaction of genetic admixture and evolutionary potential in plants and animals (Abbott et al., 2013).
Acknowledgements
We are grateful to Nicole Thomsen, Thomke Brüning, Isabel Moreau, and Sarah Frehse for assistance during experiments and animal care-taking and J. Altmüller and C. Becker for sequencing. Arne Nolte, Rafik Neme, and Jun Wang provided helpful discussions and advice for data analysis. Thanks to Katie Hitchcock and two anonymous reviewers for helpful comments to improve this manuscript. Many thanks to the numerous arachnologists, who provided samples and locality data for our study. This work is part of HKs PhD thesis. HK was supported by PhD scholarships from the Studienstiftung des Deutschen Volkes and the Max Planck Society. The study was funded by institutional funds of the Max Planck Society. The authors have no conflict of interest to declare.
Data accessibility
- FST and Pi values for all populations: FST_Pi_All_Populations.xlsx.
- Input and results files from the population modeling: Population_Modelling.rar.
- Blast results and gene expression analysis: Transcriptome_Blast_Results_Read_Counts.xlsx.
- Cold tolerance experiment results: Cold_shock_recovery.xlsx.
- Transcriptome assembly: Argiope_bruennichi_Transcriptome_Assembly.fasta.
- Genome assembly: Argiope_bruennichi_Genome_Assembly.fasta.
- Raw reads for DNA and RNA.
