Rapid Evolution of Host Repertoire and Geographic Range in a Young and Diverse Genus of Montane Butterflies

Abstract

Evolutionary changes in geographic distribution and larval host plants may promote the rapid diversification of montane insects, but this scenario has been rarely investigated. We studied the rapid radiation of the butterfly genus Colias, which has diversified in mountain ecosystems in Eurasia, Africa, and the Americas. Based on a data set of 150 nuclear protein-coding genetic loci and mitochondrial genomes, we constructed a time-calibrated phylogenetic tree of Colias species with broad taxon sampling. We then inferred their ancestral geographic ranges, historical diversification rates, and the evolution of host use. We found that the most recent common ancestor of Colias was likely geographically widespread and originated ~3.5 Ma. The group subsequently diversified in different regions across the world, often in tandem with geographic expansion events. No aspect of elevation was found to have a direct effect on diversification. The genus underwent a burst of diversification soon after the divergence of the Neotropical lineage, followed by an exponential decline in diversification rate toward the present. The ancestral host repertoire included the legume genera Astragalus and Trifolium but later expanded to include a wide range of Fabaceae genera and plants in more distantly related families, punctuated with periods of host range expansion and contraction. We suggest that the widespread distribution of the ancestor of all extant Colias lineages set the stage for diversification by isolation of populations that locally adapted to the various different environments they encountered, including different host plants. In this scenario, elevation is not the main driver but might have accelerated diversification by isolating populations.

Mountain ecosystems are home to one-third of terrestrial species (Spehn et al. 2011), with numerous rapidly radiating montane taxa (hereafter called RRMT) such as the American legume Astragalus (Scherson et al. 2008), Oreocharis (Gesneriaceae) in the Hengduan Mountains (Kong et al. 2022), and Andean Scytalopus birds (Cadena et al. 2020). It is thought that many montane radiations evolved recently (Linder 2008). Thus, studying RRMT could inform a global conundrum: What causes the rich diversity of montane species (Rahbek et al. 2019a, 2019b)? Research on RRMT has attracted the attention of an increasing number of researchers (Fjeldså et al. 2012; Hughes and Atchison 2015), but plants (Lagomarsino et al. 2016) and vertebrates (Jetz et al. 2012) have received disproportionate attention. Although some studies have been focused on understanding the diversification of montane insects (Casner and Pyrcz 2010; Ye et al. 2016; Chazot et al. 2019b; He et al. 2023; de Moraes Magaldi et al. 2024; Wang et al. 2024), a consensus has still not been reached on what abiotic and biotic factors have determined rapid global montane radiation on insects.

Changes in host repertoire (i.e., the set of hosts used by a parasite, including herbivorous plant parasites), especially changes in the number of hosts used (often referred to as host range or host breadth) seem to have an important role in insect diversification (Braga et al. 2018, 2021; Braga and Janz 2021). One of the predominant hypotheses of host-associated diversification, the “oscillation hypothesis” proposed by Janz and Nylin (2008), suggests that expansions of the host repertoire may facilitate a larger geographic distribution, followed by local adaptation, host (re-) specialization, and speciation. In other words, the diversification of insects may be increased by evolutionary changes in host repertoire and geographic distribution. However, this scenario has been rarely tested, particularly in an RRMT.

The butterfly genus Colias is an ideal taxon for studying the rapid diversification of widely distributed montane insects. Most of the nearly 80 species in this genus are distributed in the mountains of Eurasia, the Americas, and Africa (Grieshuber et al. 2012; Wu and Hsu 2017), and it is also one of the fastest diversifying genera in the family Pieridae (Carvalho et al. 2024). A reliable phylogenetic framework of Colias is necessary to understand the rapid diversification of the genus and the role that host plants might have played. Most previous studies of this genus focused on descriptions of new taxa and taxonomic reviews based on morphological traits (Verhulst 2000; Grieshuber et al. 2012; Grieshuber 2014; Wu and Hsu 2017). Although some molecular phylogenetic studies have examined Colias evolution (Pollock et al. 1998; Wheat and Watt 2008; Laiho and Stahls 2013; Kir’yanov 2021; Tunström et al. 2023), phylogenetic relationships within this genus remain poorly resolved due to inadequate species sampling or insufficient molecular markers.

Molecular markers derived from next-generation sequencing (NGS) are an established tool for reconstructing robust phylogenetic hypotheses. Many reduced representation methods can be used with NGS (Nunes et al. 2022), including target capture, which is useful for reconstructing phylogenetic relationships in rapidly radiating taxa (Larridon et al. 2020; Thomas et al. 2021). In a solution of fragmented DNA, target capture traps genomic regions of interest by hybridization with specific oligonucleotide probes, allowing non-target genomic regions to be removed (Albert et al. 2007; Hodges et al. 2007; Gnirke et al. 2009; Mamanova et al. 2010). Because the method needs relatively small amounts of DNA (Faircloth et al. 2015), it is suitable for small insect species including those with poor DNA quality represented only by specimens in museum collections (Lemmon and Lemmon 2013; Faircloth et al. 2015; Glenn and Faircloth 2016; Nunes et al. 2022). In recent years, several Lepidoptera probe sets have been developed (Breinholt et al. 2018; Zhang et al. 2019; Tian et al. 2023) for resolving the deep relationships among families (Zhang et al. 2019; Kawahara et al. 2023) and inferring phylogenetic relationships below the genus level (Kawahara et al. 2018; Ma et al. 2020).

In this study, we employed target capture with the probe sets of Zhang et al. (2019) and Tian et al. (2023) to sequence 150 nuclear gene loci and mitochondrial genomes to reconstruct the evolutionary history of Colias. Then, the divergence time of the genus was estimated using secondary calibration points. This time-calibrated molecular phylogeny was used to reconstruct historical features of this genus including its ancestral distribution, diversification rates, biogeographic history, and the evolution of host plant use. We also inferred species trees and species networks to explore putative hybridization events within the major clades of Colias. Combining our phylogenetic results with a Bayesian method for ancestral character reconstruction of host plant breadth, we infer that host repertoire evolution accelerated the rapid diversification of the genus Colias butterflies.

Materials And Methods

Taxon Sampling, DNA Extraction, and Library Preparation

We sequenced 74 samples of 66 Colias species (Supplementary Table S1). This taxon sampling includes 84% of valid Colias species with representatives from each geographic region inhabited by the genus. With the exception of C. ponteni, which is morphologically distinctive but presumed extinct, we sequenced specimens of each morphotype; the missing taxa are unlikely to have a significant impact on the topology of phylogenetic relationships (Supplementary Table S2). Zerene cesonia, in the genus most closely related to Colias (Kawahara et al. 2023), was used as an outgroup taxon. The complete genome data of Z. cesonia (Rodriguez-Caro et al. 2020; GenBank assembly accession: GCA_012273895.2) was downloaded, and all relevant sequences of this species were extracted from its genome.

Genomic DNA was extracted from the legs, thoracic, or abdominal tissues of each sample using the TIANamp Genomic DNA Kit (Tiangen, Guangzhou, China), and the extracted DNA was stored at −20 °C. DNA concentration was quantitated with a NanoDrop™ spectrophotometer. A volume of 100 ng of genomic DNA was sheared using a Scientz18-A ultrasonic processor (SCIENTZ, Zhejiang Province, China) to a size of 300–550 bp. We prepared libraries with the NEBNext Ultra DNA Library Prep Kit (New England Biolabs, Ipswich, MA, USA). Each library was labeled with a unique 7-bp P7 index sequence. Four or 5 libraries were mixed into a pooled library in equal concentrations for subsequent hybrid capture, and a total of 15 pooled libraries were generated.

Probe Preparation, Hybridization, and Sequencing

We used the target capture strategy described by Zhang et al. (2019) and Tian et al. (2023) to collect genetic data. We targeted 150 nuclear protein-coding (NPC) loci of Lepidoptera. The PCR primers for these 150 NPC loci are provided in Supplementary Table S3. We prepared amplicon capture baits for our target capture experiment following the method of Zhang et al. (2019). In brief, we mixed 20 ng of genomic DNA from each of the 74 butterfly samples and this pooled DNA was then used as a template to amplify the 150 NPC loci via polymerase chain interaction (PCR). The 150 PCR products were then mixed together and purified with AMPure XP beads (Beckman Coulter, Brea, CA, USA). The purified amplicon mixture was subjected to end-repair and A-tailing, and then ligated with a biotinylated Bio-T adapter (Zhang et al. 2019). The adapter-ligated amplicons were purified with AMPure XP beads and subsequently immobilized on Dynabeads MyOne streptavidin magnetic beads (Life Technologies, Carlsbad, CA, USA) to obtain bait-coated beads. Detailed experimental protocols are provided by Zhang et al. (2019).

For each hybridization reaction, 500 ng of pooled library and 2.5 μl of bait-coated beads (containing 25 ng of biotinylated amplicons) were used. A touch-down hybridization program was adopted. After denaturation, the hybridization started at 65 °C, decreased by 5 °C every 6 h, and ended at 45 °C, for a total duration of 30 h. The enriched libraries were amplified using Illumina P5 and P7 primers and purified with AMPure XP beads. Finally, all 15 enriched libraries were pooled in equal concentrations and sequenced on an Illumina HiSeqX10 sequencer in the paired-end 150-bp mode. In addition, we repeated the hybridization and sequencing of samples with unsatisfactory initial sequencing results.

Data Processing and Data Set Assembly

Bases were identified from raw image files with bcl2fastq2 (https://emea.support.illumina.com/downloads/bcl2fastq-conversion-software-v2-20.html) to generate raw reads. These sequences were then subjected to quality control and filtering by removing adapter sequences and low-quality reads with more than 5% N bases. Reads were then de-multiplexed and screened with Trimmomatic v0.33 (Bolger et al. 2014), FastQC v0.11.9 (http://www.bioinformatics.babraham.ac.uk/projects/fastqc/), and FastUniq v1.1 (Xu et al. 2012). MitoFinder v1.4 (Allio et al. 2020) was used to assemble mitochondrial genomes and each of the 150 targeted nuclear loci using the cleaned reads and the assembled outgroup taxon genome sequences (Rodriguez-Caro et al. 2020). Cd-hit-est v4.0 (Li and Godzik 2006) was used to further filter contigs to reduce redundancy. Then, SAMtools v1.4.1 (Li et al. 2009) was used to calculate the sequencing depth, and sequences of target loci with an average sequencing depth ≥ 5× were retained for further analysis. tBLASTN v2.2.18 (Altschul et al. 1990) was used to identify orthologous sequences from the contigs of each locus in each sample; sequences of our 150 target loci were retrieved from the Danaus plexippus genome as references (GenBank assembly accession: GCA_009731565.1). Potential chimeric sequences were identified by homology comparison and removed. Flanking sequences were identified using reference sequences. The sequences of each targeted locus were then obtained by taking the consensus sequence of each contig of sequencing reads after all quality control steps. MAFFT v7.0.1 (Katoh and Standley 2013) was used to align reads by comparing them to the 150 reference sequences that were used to design the probe set, and the process was optimized using Gblock v0.91B (Castresana 2000). Sequences corresponding to abnormally long branches were deleted. Two phylogenomic data sets were finally made for our samples: 1 was a nuDNA data set of 150 NPC loci, and the other was a mtDNA data set of mitochondrial genomes.

Phylogenetic Analysis

The reading frame of each nuclear and mitochondrial locus was determined using Exonerate v2.2.0 (Slater and Birney 2005). The data sets were partitioned by gene and codon positions, except that the non-protein coding mitochondrial 16S rRNA, 12S rRNA, and tRNA genes were each considered to be single, independent partitions. For each of the nuDNA and mtDNA data sets, the maximum likelihood (ML) gene tree was inferred with IQ-TREE v2.2.0 (Minh et al. 2020a). The best model and partition scheme were calculated using the ModelFinder (Kalyaanamoorthy et al. 2017) in IQ-TREE, with the predefined codon partitions Model Finder Plus (MFP) with the MERGE option. The options (-allnni -nstop 500 -pers 0.5) were employed to achieve a more comprehensive tree search. Node supports were assessed by performing the standard bootstrap test with 500 replicates (-b 500), SH-like approximate likelihood ratio test (SH-aLRT) with 5000 replicates (-alrt 5000), and ultrafast bootstrapping with 5000 replicates (-B 5000). ML support values are regarded to be high when bootstrap support (BS) is ≥ 70% (Hillis and Bull 1993), ultrafast bootstrap values are ≥ 95% (Minh et al. 2013), and SH-aLRT is ≥ 80% (Guindon et al. 2010).

The species trees of nuDNA were inferred using 2 different schemes. For the first scheme, all 150 nuDNA loci were separately used to infer the unrooted gene trees in IQ-TREE v2.0.5 (Minh et al. 2020b). The species tree was reconstructed from the 150 unrooted gene trees in Asteroid v1.0, a method to infer species trees from gene trees using data sets with substantial missing loci (Morel et al. 2023). Bootstrap values were computed from the input gene trees with 50,000 replicates. For the second scheme, a strict filtering criterion applied to the nuDNA data set resulted in a data set of 32 single-gene nuclear loci with all samples, which were used to infer gene trees in IQ-TREE v2.0.5. The species tree was built using the gene trees by ASTRAL-III v5.6.1 (Zhang et al. 2018) with posterior probabilities of local quartet topology (LPP), and the concordance factors (gene concordance factor/sequence concordance factor, Minh et al. 2020a) were calculated comparing the trees by IQ-TREE v2.0.5. With regard to the mtDNA species tree, 13 protein-coding and 2 tRNA loci were selected to infer the gene trees using RAXML v8.0 (Stamatakis 2014) with 200 bootstrap replicates, as the tRNA sequences were too short for gene tree reconstruction. The final species tree of mtDNA was obtained for these gene trees with ASTRAL-III v5.6.1.

Hybridization Inference

Preliminary analyses of our genetic data suggested that introgression between Colias lineages was common. This causes portions of the genomes of two lineages to be similar because of hybridization, not from descent from a recent common ancestor. Species Networks applying Quartets (SNaQ; Solís-Lemus and Ané 2016), implemented in PhyloNetworks (Solís-Lemus et al. 2017), was used to infer the phylogenetic network of each nuclear locus in a maximum pseudolikelihood framework. This method has been recently applied to phylogenetic network analysis of data sets with a large amount of missing data (Blair et al. 2019; Obiol et al. 2021; Calderón-Acevedo et al. 2022). The analysis time of this method increases drastically when there are more than 25 terminal taxa in the data set (Hejase and Liu 2016). To circumvent this limitation, the full data set of 66 unique Colias species was split into 5 data subsets. Each included the outgroup Zerene cesonia and one sample of each species in each of the other major clades in the species tree inferred with Asteroid v1.0. This method can identify hybridization events within each clade, but not between clades (except for hybridization between those taxa and the outgroup). Because the largest clade had 24 species, it was not possible to analyze 2 clades in a single analysis as most of the data sets would surpass the limit of 25 taxa. For each of the 5 data subsets, SNaQ was run with a set of gene trees inferred with IQ-TREE and a starting species tree generated with Asteroid v1.0. For each of the 5 clades, we specified the maximum number of hybridization events (hmax) in SNaQ using values ranging from 0 to 5. The pseudo-likelihoods for each hmax were plotted to select the most likely hmax for each clade. All 5 phylogenetic networks were reassembled into a single tree with the ape R package (Paradis and Schliep 2019).

Divergence Time Estimation

The nuDNA species tree inferred with Asteroid v1.0 was used as the fixed topology for estimating divergence times. In the absence of any fossils that can confidently be assigned to the subfamily Coliadinae (de Jong 2017), we used secondary calibrations to date the tree. The divergence time of each node was estimated with the MCMCTree program in the PAML package (Yang 2007). Two calibration points from previous studies were used: one is 12.13 Ma (10.6–15.71 Ma) for the common ancestor of Colias + Zerene from Kawahara et al. (2023) and the crown age of Colias, which was set as a range based on the minimum and maximum divergence dates from 2 different studies. Ranging from 1.83 Ma (Tunström et al. 2023; Colias crown age estimate 1.83-3.02 Ma) to 6.4 Ma (Carvalho et al. 2024; Colias crown age estimate 4.21–6.4 Ma), which covers the minimum and maximum crown ages of Colias estimated in recent investigations. We adopted the lognormal relaxed clock and General Time Reversible (GTR) + Gamma (G) model with other parameters as defaults. The analysis was run for 15 million generations, sampled every 150 generations with 1.5 million generations discarded as burn-in. The convergence was evaluated in Tracer v1.7 (Rambaut et al. 2018), with effective sample sizes larger than 200 for all parameters.

Ancestral Area Reconstruction

To reconstruct the biogeographic history of the genus Colias, we compiled distribution data for 79 Colias species from the literature (Supplementary Table 4). We then assigned the 66 Colias species in the time-calibrated tree (excluding only Colias hecla viluiensis) to one or a combination of 7 biogeographic regions: Neotropics, Nearctic, Southeast Asia + India, East Palearctic, West Palearctic, Afrotropics, and Pan-Tibetan Highlands. We performed Bayesian inference of historical biogeography using the Multi-feature Feature-Informed GeoSSE (MultiFIG) model (Swiston and Landis 2023). This model allows the estimation of biogeographic processes such as dispersal, extinction, and speciation, in addition to spatial and environmental features of regions that are expected to impact one or several of these processes. We first delimited polygons representing each biogeographic region (following Chazot et al. 2021; Liu and Zhu 2022; Liu et al. 2022). Then, for each region, we retrieved geographic features related to surface area, altitude, and distance to other regions, using the R package regionfeatures (https://bitbucket.org/sswiston/regionfeatures). We extracted data for 8 regional features: quantitative area, categorical area (above/below mean of all regions), quantitative distance (mean distance between a randomly selected point in the first region and a randomly selected point in the second region), categorical distance (adjacent/non-adjacent), quantitative altitude (mean altitude), categorical altitude (above/below mean), quantitative altitudinal difference (difference in mean altitude), and categorical altitudinal difference (whether regions share highland/lowland status or not).

We performed an analysis in RevBayes (Höhna et al. 2016) to infer base rates of dispersal, extinction, within and between-region speciation, the feature effect parameters, and ancestral biogeographical ranges. To tune hyperparameters, we ran 2 independent runs of 50,000 generations, sampling parameters and node histories every 100 cycles, and discarding the first 10% as burn-in. To verify that Markov chain Monte Carlo (MCMC) analyses converged to the same posterior distribution, we applied the Gelman diagnostic (Gelman and Rubin 1992) as implemented in the R package coda (Plummer et al. 2006), with a threshold of 1.1 to confirm convergence. Results presented here are drawn from 1 MCMC run.

Diversification Rate Analysis

To estimate palaeodiversity dynamics and test for heterogeneity of diversification rates across clades within Colias, we used the clade-shift model of Morlon et al. (2011) as implemented in the R package RPANDA (Morlon et al. 2016; Mazet et al. 2023). We tested diversification shifts at the crown age of 3 clades determined by their geographic distribution: (i) the ancestor of the 7 Neotropical species included in our tree; (ii) the ancestor of the sister group to the Neotropical clade, which includes all species in the Northern Hemisphere; and (iii) the ancestor of all species that occur in the Nearctic region. We used the compiled data set of the distribution of 79 Colias species (Supplementary Table S4) as a reference for calculating the sampling fraction for each clade. By doing this, we assumed that unsampled species that occur anywhere but the Neotropics are in Clade B (the non-South American clade) and that unsampled species that occur in the Nearctic are within the Nearctic subclade.

After identifying the best combination of shifts, birth-death models, and parameter values, we tested model adequacy by simulating 3351 trees (from 10,000 attempts) under the best-fit model combination. Following Mazet et al. (2023), we compared simulated trees to the empirical tree in terms of number of lineages through time, species richness per clade, and average leaf depth index, which is a measure of tree imbalance.

Even though this analysis was informed by geographical distribution (for clade selection), diversification rates estimated by the clade-shift model and MultiFIG are not directly comparable because these methods model different types of rate heterogeneity. MultiFIG estimates diversification rates associated with each geographical range and whether spatial and environmental features can explain differences in rates between ranges. Conversely, the clade-shift model infers diversification rates change over time based only on the phylogenetic tree.

Additionally, we tested for elevation-dependent speciation using ES-sim, a simulation-based test that uses a tip-specific metric of speciation rate derived from branch lengths. This approach is similar to the quantitative state speciation and extinction (QuaSSE) method in terms of statistical power to detect trait-dependent diversification and is more robust against false positives (Harvey and Rabosky, 2018). We implemented ES-sim using the R script available at https://github.com/mgharvey/ES-sim, using elevation data for each Colias species (Supplementary Table S4, Supplementary Fig. S1).

Host Repertoire Evolution

To infer the evolutionary history of associations between Colias butterflies and their host plants, we performed Bayesian inference under the 2-state model of host repertoire evolution proposed by Braga et al. (2020). Given a phylogenetic tree for each interacting clade and a matrix containing information about which ecological interactions have been observed between extant lineages, this method models the gains and losses of specific host taxa along the symbiont phylogeny (in our case, the butterflies) that resulted in the observed interactions. The host phylogeny is used to infer whether hosts that are closely related to other hosts already consumed by the species are more easily gained compared to more distantly related hosts. Preliminary analyses demonstrated that phylogenetic distance between host plant genera does not affect host gain probabilities in Colias. Thus, the analysis presented here does not include the host plant phylogeny.

We used information gathered from the literature to compile an interaction matrix between 55 Colias species and 62 host plant genera (Supplementary Table S5). For the butterflies, we used the reconstructed time-calibrated Colias tree and kept only the 55 species with host use data. The joint posterior distribution of model parameters and evolutionary histories were estimated using the inference strategy described in Braga et al. (2020) as implemented in RevBayes (Höhna et al. 2016). We ran 3 independent MCMC analyses for 10⁵ cycles, sampling every 100 cycles. The first 20–50% of samples were discarded as burn-in; burn-in was determined by visual inspection of ESS and likelihood values in Tracer (Rambaut et al. 2018); and then 500 random samples were taken from each independent run. We used the implementation of the Gelman diagnostic (Gelman and Rubin 1992) in the R package coda (Plummer et al. 2006) to verify that MCMC analyses converged to the same posterior distribution. We present results from all MCMC analyses combined.

The posterior distribution of ancestral interactions was summarized in 2 ways using the R package evolnets (www.github.com/maribraga/evolnets): (i) interactions with at least 90% posterior probability at the internal nodes of the butterfly tree; and (ii) weighted summary networks at 3, 2, and 1 Ma where interactions with at least 70% posterior probability are included. This probability was used as a weight for the link between nodes. To facilitate visualization and evaluate whether interactions with groups of hosts tend to evolve together, we identified modules in the extant network and the 3 ancestral networks. We then validated the ancestral modules by calculating how often 2 nodes were placed in the same module across 100 networks sampled during MCMC. For the extant network, in addition to modularity (Beckett 2016) we also calculated nestedness using the Nestedness metric based on Overlap and Decreasing Fill (NODF; Almeida‐Neto et al. 2008; Almeida-Neto and Ulrich 2011) and compared it to 100 networks generated by a null model that maintains the number of interactions in the observed network. This comparison was done by calculating the z-score, which quantifies the position of the observed metric within the null distribution in standard units, revealing how much it deviates from the null expectation (Ulrich et al. 2009). All data, RevBayes and R scripts for the analyses of historical biogeography, clade-shift diversification, and host repertoire evolution are available at www.github.com/maribraga/Colias_hostrep and the Dryad repository https://doi.org/10.5061/dryad.bk3j9kdkz.

Results

Sequence Capture and Data Matrix Statistics

The success of hybrid capture and Illumina sequencing, including attributes of the 150 nuclear loci we targeted, is summarized in Supplementary Table S6. A total of 235,881,658 clean paired-end reads were obtained; the capture success of each sample is summarized in Supplementary Table S7. The number of clean reads per sample ranged from 582,057 to 11,680,568, with an average of 3,187,590 reads per sample. In comparison with the reference sequences of the targeted loci, the percentage of missing sequences for each sample ranged from 2.92% to 74.86%, with an average of 20.48%. The read‐to‐target mapping percentage (on‐target) of the samples with secondary hybridization was significantly higher than those without secondary hybridization. The average sequence coverage of the orthologous contigs was 397×, with a wide range from 1 to 7419× (Supplementary Table S8). The general features of the mitochondrial genome data set are summarized in Supplementary Table S9. Two concatenated matrices were used for phylogenetic analysis: the 150 nuDNA locus data set was 179,022 bp in length and the mitochondrial genome data set was 12,604 bp in length. GenBank accession numbers of the newly sequenced markers were summarized in Supplementary Table S10.

Phylogenetic Analyses

Two characteristics can be seen in all trees: (i) 2 main clades of Colias, hereafter called Clade A and Clade B, were recovered with strong support; and (ii) some species represented by multiple samples, including C. croceus, C. erate, and C. hecla, were not recovered as monophyletic.

The Asteroid species tree of nuDNA (Fig. 1) demonstrates that the genus Colias is monophyletic with full support (BS = 100). The monophyly of Clade A is fully supported (BS = 100). Within this clade, the sister to the remaining South American taxa is C. dimera, which is distributed in northern South America and has a unique male wing color. The monophyly of Clade B is also fully supported (BS = 100), and, while intermediate nodes lack support, stronger node support is found among numerous closely related taxa. Some species pairs were recovered as monophyletic with strong bootstrap values, including C. fieldii and C. tibetana; C. aurorina and C. chlorocoma; and C. diva and C. heos. The highly supported clades of the ASTRAL species tree (Supplementary Fig. S2) are almost identical to those of the Asteroid species tree (Fig. 1).

Figure 1.

Comparison between 2 Colias trees with representative adult photos of Colias upper sides. Left) A species tree inferred with Asteroid using gene trees of 150 nuclear gene loci. BS is denoted at the node based on 50,000 replications. Right) Reassembled phylogenetic network inferred using 5 SNaQ analyses, one for each clade. The arrows indicate the direction of the hybridization. The numbers near the arrows represent γ values, indicating the proportion of the genome inherited via introgression; larger values of each pair are green. Arrows with solid lines indicate introgression within a clade, and arrows with dashed lines indicate hypothesized introgression between 2 different clades.

The topology of the nuDNA gene tree (Supplementary Fig. S3) differs from the nuDNA species trees (Fig. 1 and Supplementary Fig. S2), but lineages with high support values are coincident among these trees. The mtDNA gene tree (Supplementary Fig. S4) had lower support for Clade A (BS = 68). All South American samples form a strongly supported monophyletic group (BS = 100), which is divided into 2 clades. One clade includes only C. vauthierii, which is found in Chile and Argentina south of the Andes and is restricted to high elevations; the other clade has short branches and is primarily distributed throughout the Andes. The monophyly of the non-South American Clade B is also strongly supported (BS = 100), but the internal branches differ greatly from the relationships in the nuDNA gene tree (Supplementary Fig. S3). The first-diverging clade is the widely distributed C. hyale clade, including C. hyale, C. alta, and C. alfacariensis. All remaining species are grouped into a monophyletic group (BS = 100) and diversified rapidly. The topology of the species tree inferred with 15 mitochondrial loci (Supplementary Fig. S5) is almost identical to the mtDNA gene tree (Supplementary Fig. S4).

Divergence Time Estimation

The genus Colias diverged from the common ancestor of Zerene + Colias in the mid-Miocene ~13.33 Ma (95% Confidence Intervals, CI: 10.63-15.74 Ma) (Supplementary Fig. S6). The common ancestor of the extant Colias species originated ~3.50 Ma (CI: 2.04–5.55 Ma). The crown ages of Clade A and Clade B are ~1.83 Ma (CI: 1.02–3.05 Ma) and ~3.07 Ma (CI: 1.80–4.89 Ma), respectively. Further diversification of this genus occurred at ~0.5–3 Ma.

Inference of Hybridization

The reassembled SNaQ phylogenetic network had a different topology than the Asteroid species tree, even though the topology of the species tree was the starting point for the SNaQ analysis (Fig. 1). Topological differences between the 2 trees occur in lineages with low BS (60% > BS), which result from discordant gene trees that likely result from hybridization. Although the node C. flaveola shares a highly supported node with its sister clade (100% > BS ≥ 90%) in the Asteroid tree, change of its phylogenetic position is found along with hybridization events in the SNaQ tree. Lineages including C. mossi, C. berylla, C. stoliczkana, C. wanda, C. arida, C. croceus, and C. interior have low BS and are inferred to have hybridized.

The 2 hybridization events in Clade 5 demonstrate the 2 types of inference that can result from our approach. The first type is introgression between taxa in 2 distinct lineages. For example, a hybridization event is inferred between C. lesbia and C. vauthierii, albeit with a non-significant value (γ ≈ 0.09). The other type of inference is introgression between a taxon and its ancestral lineage. For instance, Colias flaveola shares nearly half its genome (γ ≈ 0.42) with descendants of its ancestral lineage including C. mossi, C. euxanthe, and C. weberbaueri. In total, we infer 14 hybridization events. Dashed lines in the species network indicate inferred hybridization events between 2 lineages from inside and outside the clade, respectively. Because of computational limitations that led us to analyze each clade separately, we were unable to infer these putative hybridization events more precisely (Fig. 1).

Reconstruction of Ancestral Areas

The MultiFIG analysis estimated that the ancestral range of Colias included East and West Palearctic + Pan-Tibetan Highlands + Nearctic + Neotropics (posterior probability = 0.41, Fig. 2). The range with the second highest posterior probability was much less likely (0.08) and included all regions except for the Afrotropics. The 8 ranges with the highest posterior probability (which together sum to 0.808) included between 4 and all 7 regions (Supplementary Tables S11 and S12). More importantly, among the regions in the most likely ancestral range, the Pan-Tibetan Highlands are included in all 8 ranges, while all other regions are included in 7 of the 8 most likely ranges. Thus, we cannot estimate the specific ancestral range with confidence. The most likely scenario suggests a widespread origin, from which different clades diversified in different regions of the world. A few lineages underwent new expansion phases, colonizing, for example, the 2 remaining regions: Southeast Asia and the Afrotropics (Fig. 2).

Figure 2.

Historical biogeography of the genus Colias. a) Ancestral range reconstruction done with MultiFIG. The size of circles at internal nodes represents the posterior probabilities for the inferred historical ranges that are indicated with color. Asterisks mark the locations where diversification shifts were tested. b) Delineation of the 7 regions used in this biogeographic analysis.

Of the 8 relationships between-region features (i.e., area, altitude, distance) and processes (i.e., speciation, extinction, dispersal) that we examined, 3 were estimated with high posterior probability when considering the joint probability of categorical and quantitative parameters (Supplementary Fig. S7). First, the model assigned a high probability (0.99) to the scenario where quantitative and/or categorical distance between regions negatively affects the dispersal rate, so that closer regions are more easily colonized. Therefore, we can conclude that distance influences dispersal even if we cannot confidently distinguish between the effects of distance (quantitative parameter) and adjacency (categorical parameter) between regions. Second, there was a high probability (0.95) that quantitative and/or categorical areas negatively affect the extinction rate, meaning that extinction is less likely in larger regions. Third, the model assigned a probability of 0.91 to the scenario where quantitative and/or categorical distance between regions negatively affects the between-region speciation rate. This negative relationship is probably explained by the higher likelihood that widespread ranges include adjacent regions. Variation in diversification rates among biogeographical ranges was mainly determined by range size, where between-region speciation was more likely in ranges including more regions (Supplementary Figs. S8 and S9). Finally, no feature related to elevation was inferred to have an effect on dispersal, speciation, or extinction (0.76 < P < 0.82).

Diversification Rate Analysis

Of the 4 possible combinations of shifts with a single backbone and the 5 combinations with multiple backbones, 2 were detected as the best combinations (ΔAICc = 0.21 between first and second-best models, Fig. 3 and Supplementary Fig. S10). The best combination includes a single shift in diversification regime at the base of Clade B while the second-best combination also includes a shift at the base of the clade of Nearctic Colias taxa. Both subclades are best explained by a model with an exponentially decreasing speciation rate and no extinction while the backbone (Clade A) is best explained by a model of diversification with constant speciation through time (Fig. 3b and Supplementary Fig. S10b). Both combinations result in an increasing species diversity through time, which slows down to the present and reaches the current diversity (Fig. 3 c and Supplementary Fig. S10c). We found no correlation between elevation and tip-diversification rates (min. elevation: ρ = −0.15, P = 0.45; max. elevation: ρ = −0.04, P = 0.80, elevational range (max–min): ρ = 0.15, P = 0.41).

Figure 3.

Diversification shifts and palaeodiversity dynamics of Colias. a) Colias most likely experienced a single shift in diversification rate, with slower diversification in South America than in the Northern Hemisphere. b) Clade-specific diversification rates over time. Only speciation rates are shown, as the best-fit models do not include extinction. c) Clade-specific species accumulation curves.

Host Repertoire Evolution

The extant network of 55 Colias species and 62 host plant genera reached the highest modularity in a configuration with 5 modules of butterflies and plants that interact more with each other than with the rest of the network (Supplementary Fig. S11a). Because the algorithm searches for the most modular network configuration without taking into account any biological information, modules are not always easy to interpret. Still, there was a general pattern. Module 2 was defined by butterflies with the broadest host repertoires (C. erate and C. eurytheme) and their close relatives, while all other modules were defined by the host plant genera. The host plants with the most interactions in each module were: Astragalus and Medicago in M1; Trifolium and Vaccinium in M3 (even though rarely used together); Oxytropis, Lupinus, and Hedysarum in M4; and Coronilla, Vicia, and Lotus in M5. Interestingly, the extant network is not more modular than expected by the null model (Q = 0.42, z-score = −1.89), but is more nested (NODF = 33.22, z-score = 49.24). Still, the origin of most modules can be traced back to one of the ancestral networks.

Our analysis of host repertoire evolution inferred 343 events (95% highest posterior density [HPD] 296–398) along the Colias tree, 43% of those host gains and 57% of host losses. This is equivalent to approximately 1.5 gain and 2 loss events per million years along each branch of the tree. The most recent common ancestor (MRCA) of all Colias species included in this study likely used Astragalus (module M1) and Trifolium (M3) as host plants (posterior probabilities are 0.99 and 1, respectively). Ancestors of Clade A used combinations of the ancestral hosts (Astragalus and Trifolium) and Medicago (M1), with other hosts being colonized more recently by individual species (Fig. 4). On the other hand, there were several events of host range expansion and contraction within the much larger Clade B. The MRCA of Clade B likely used Oxytropis (M4), in addition to the 2 ancestral hosts (Fig. 4). In the subclades of Clade B, most lineages lost some part of the ancestral host repertoire, except for 1 clade (C. electo and close relatives) that greatly expanded its host range, colonizing hosts from all 5 modules. One of the extant butterflies (C. electo) in this clade (C. electo and close relatives) then dispersed to Africa, while all other butterflies remained in Eurasia. In the clade of Nearctic Colias taxa, several host range expansions were reconstructed. Most of them added 1 or 2 hosts, but larger host range expansions happened in the MRCA of C. alexandra and C. philodice and its descendants. Two lineages went in the opposite direction and reduced their host repertoires to a single host, Vaccinium (M3). Their extant descendants use 1 or 2 host genera in addition to Vaccinium.

Figure 4.

Evolution of host repertoire in Colias butterflies. a) Asteroid Species Tree with ancestral repertoires at internal nodes with at least 90% posterior probability. Each square represents one host genus and has the same color as in b. b) Interaction network between Colias species and their host plants (see Supplementary figures for taxon names). Interactions are colored by modules, which are groups of butterflies and hosts that interact mainly with each other. Gray squares represent interactions between different modules.

We also reconstructed ancestral butterfly-plant networks at 3, 2, and 1 Ma (Supplementary Fig. S11b–d). We identified modules from summary networks (networks, where interaction weight is given by posterior probability and only interactions with at least 0.7 probability are included) and validated them with networks sampled during MCMC (Supplementary Table S13). At 3 Ma there was only 1 module, including the 3 Colias lineages extant at that time and 3 host plants: Astragalus, Trifolium, and Oxytropis. At 2 Ma, a new module was formed. It included the 3 earliest lineages in the clade of Nearctic Colias taxa and the host plants Hedysarum, Lupinus, and Vaccinium, the first non-Fabaceae plant colonized by Colias. At 1 Ma, M4 expanded and gave origin to M3, which included the Vaccinium specialists. Until this point in time, Trifolium was placed in M1, only switching to M3 in the extant network. This change does not seem to have any biological implications and is simply a stochastic result of the modularity algorithm. Finally, also at 1 Ma, module M5 originated by a host range expansion at the branch leading to the ancestor of C. electo and its sister clade.

Discussion

Rapidly radiating taxa have always posed challenges for inferring phylogenetic relationships with confidence (Larridon et al. 2020). The development of reduced representation NGS methods has enabled phylogenetic reconstruction using hundreds of genetic loci, making evolutionary inferences of recent and ancient divergences more reliable (Grover et al. 2012; Lemmon and Lemmon 2013; Nunes et al. 2022). A robust global phylogeny of Colias has not been recovered in previous studies, largely resulting from the limited genetic data (Kir’yanov 2021) or inadequate taxon sampling (Tunström et al. 2023; Carvalho et al. 2024). Disentangling the rapidly evolving history of geographic range and host repertoire in the montane genus Colias was our motivation for employing a probe set of 150 nuclear loci with mitochondrial genomes to sequence > 190,000 bp/specimen from the most extensive taxon sampling of Colias yet assembled.

Time-calibrated Phylogeny and Species Network

Different phylogenetic analyses of our data all find 2 well-supported main clades within Colias: Clade A includes South American taxa, and Clade B includes non-South American taxa. Hybridization is inferred to have occurred within Clade A and within Clade B, but not between them, as would be expected for these geographically separated lineages. This result further supports the 2-clade division of Tunström et al. (2023). Several lines of evidence suggest that the diversification of Colias was rapid and recent. We infer the Colias crown group to be just ~3.5 Ma old, and introgression between lineages seems to have been rampant, as is often observed in young radiations (Fig. 1; Tunström et al. 2023). The recency of divergence and the frequency of hybridization pose a challenge for phylogenetic inference because of incomplete lineage sorting and an evolutionary history that is not strictly bifurcating. Some species with more than 1 individual in our taxon sample (such as C. croceus, C. erate, and C. hecla) were not monophyletic, which suggests cryptic species-level diversity or frequent, ongoing gene flow. We infer hybridization between C. electo and the lineage that gave rise to C. croceus, and these 2 species are known to hybridize (Berger 1986; Descimon and Mallet 2009). Strong discordance between the trees of mitochondrial genomes (Supplementary Fig. S5) and nuclear loci (Fig. 1) is further evidence of frequent introgression among lineages. Accompanied by hybridization events, the topological uncertainty of many nodes is found in our Colias trees, it should be noted that the presentation of our results focuses mainly on the supported nodes.

Our divergence time estimation of the stem and crown ages of Colias (Supplementary Fig. S6) yields similar results to those of recent studies (Chazot et al. 2019a; Kawahara et al. 2023; Tunström et al. 2023; Carvalho et al. 2024). An older stem age of ~20.1 Ma (10.4–32.3 Ma) was suggested for the initial split of Colias from the MRCA of its sister taxon Zerene, based on a tribal-level dated phylogeny on which only 1 sample was used for each genus of Colias and Zerene (Espeland et al. 2018). This old age is probably caused by the sparce taxon sampling at the genus and species levels.

Diversification of Colias

We found that Clade A has a constant speciation rate, and Clade B has an exponentially decreasing rate in which diversification is higher in the past and exponentially decreases towards the present (Fig. 3). This latter pattern is considered by some to be the main feature of adaptive radiation (Gavrilets and Losos 2009; Moen and Morlon 2014). In addition, Carvalho et al. (2024) found that Colias is one of the most rapidly diversifying lineages within the family Pieridae as a whole. The burst of early diversification may result from recently diverged species filling unoccupied niches after entering new adaptive areas (Kong et al. 2022) until the newly available niche space becomes saturated (Yoder et al. 2010), and the diversification rate begins to slow (Phillimore and Price 2008; Rabosky 2013).

Many factors may lead to niche shifts, including intrinsic factors such as key innovations, hybridization, introgression, and species interactions (López-Fernández et al. 2010; Rabosky and Glor 2010). Extrinsic factors including geologic or (warming) climatic events (Rohde 1992; Carvalho et al. 2024) and mountain uplift (Lagomarsino et al. 2016; Esquerré et al. 2019) may also cause niche shifts. Diversification cannot be explained simply by examining diversification rates, so we will discuss Colias diversification in concert with results from our analyses of historical biogeography and host plant repertoire.

Historical Biogeography

Diversification across different elevations has been considered one of the driving forces in the RRMT (Steinbauer et al. 2013; Halbritter et al. 2018), and likely triggered the diversification of many biological taxa. Even though Colias occurs in montane areas of varying elevation (Supplementary Table S4, Supplementary Fig. S1), our tip rate analysis and the estimates of the relationships between region features and biogeographical processes indicate that elevation has not affected dispersal, speciation, or extinction, which means that elevation may have no effect on the rapid diversification of Colias.

Montane species are profoundly affected by climate change, which causes changes in population size (demography) as well as latitudinal and altitudinal distributions (Hewitt 2000; Vintsek et al. 2022). Most Colias species diversified in the Northern Hemisphere during the past ~3.5 Ma, and geographic expansion events of species in the Western Palearctic were common (Fig. 2). It may be a pattern that repeats with each glacial cycle: (i) During glacial periods, cold-adapted montane species may spread into lowland areas and laterally into other mountain areas (Wallis et al. 2016). (ii) During interglacial periods, their ranges may shrink, isolating populations on mountain peaks (Minter et al. 2020). This distribution pattern associated with glacial cycles may produce genetically isolated populations, which may further promote speciation after the newly separated populations adapt to their isolated mountain homes. The geographic expansions of Colias may also be aided by evolutionary changes in host repertoire and the use of widespread host plants.

Host Repertoire Evolution

Host repertoire evolution in Colias butterflies (Fig. 4) was characterized by many host range expansion and contraction events. From an ancestral host repertoire including the legume genera Astragalus and Trifolium, Colias butterflies colonized a wide range of Fabaceae genera as well as plants in more distantly related families. This variation is mainly seen in Clade B, as extant species in the South American Clade A still only use the ancestral hosts and a few other closely related genera. Among extant species, there are 2 phylogenetically independent groups of more generalist Colias taxa that have colonized many Fabaceae genera (often the same ones, suggesting that they live in similar habitats), including early-diverging Fabaceae genera within this diverse family. While one of these cases can be confidently explained as an extensive host range expansion event by the ancestor of C. philodice and C. eurytheme, the timing of the other event (in the subclade including C. erate, Clade 2) is less clear given the poor branch support for this subclade (Fig. 1), likely caused by high levels of gene flow within it. Still, generalism is seen at the tips of the phylogeny, and generalist species are often closely related to specialist species, which is a general pattern supporting the oscillation hypothesis (Janz and Nylin 2008). This pattern, which is found in phytophagous insects and other parasites (Nylin et al. 2018), demonstrates that species sometimes become more generalist, but then typically specialize again so that only recent host range expansions are observed.

The nestedness observed in the extant butterfly-plant network (with ancestral hosts being the plants with the most interactions) is another line of evidence that oscillations in the host range have occurred across the Colias tree (Braga et al. 2018). The low network modularity and lack of modules with many closely related butterflies on a single host suggest that the interactions between Colias butterflies and their host plant genera cannot be explained by explosive radiations on novel hosts (Braga et al. 2018). These results seem to indicate that the oscillation hypothesis (Janz and Nylin 2008) can operate at various taxonomic and temporal scales, while the escape-and-radiate hypothesis (Ehrlich and Raven 1964) might operate mainly at the host plant family level.

By combining results from the analyses of historical biogeography and host repertoire evolution, we can investigate the role of the host plants in the butterflies’ dispersal across the world. The origin and early diversification of Astragalus likely happened in Asia more than 10 Ma (Azani et al. 2019), and some colonization of the New World likely happened before Colias diversification (~3.5 Ma). Thus, the widespread distribution of Astragalus seems to have facilitated the dispersal of Colias from South America to Asia (likely through North America) and subsequent diversification. Another example of the combined evolution of geographic and host ranges is the dispersal back to North America in one of the subclades of Clade B. Even though phylogenetic uncertainty prevents more confident timing of colonization of specific host genera, early species in this subclade have gradually added new genera into their host repertoires (including genera from distantly related families), which likely facilitated geographic range expansion. One of these genera is Vaccinium, which is mainly found in cooler areas of the Northern Hemisphere (Seider et al. 2022; Hirabayashi et al. 2023). Thus, butterflies could use Vaccinium in North America and Asia, likely helping species achieve widespread distributions, in some cases followed by local adaptation and speciation.

Conclusions

We present a well-sampled, dated phylogeny of Colias and show evidence for recent, rapid radiation during its short evolutionary history. This genus originated from a widespread ancestral range and split into the South American and non-South American clades ~3.5 Ma. Lineages of Colias diversified in many different regions across the world as they expanded geographically, and, surprisingly, we conclude that elevation probably does not affect their diversification. The genus experienced a burst of diversification in its early evolutionary history, followed by an exponential decline in diversification rate until the present. From an ancestral host repertoire including the legume genera Astragalus and Trifolium, the host ranges of Colias butterflies contracted and expanded to include a wide range of Fabaceae genera and plants in more distantly related families. We suggest that the diversity of Colias was facilitated by its geographical expansion and adaptive radiation into new niches, aided by the colonization of new host plant taxa. Our study highlights that the species richness of rapidly radiating montane insect taxa may not necessarily be related to altitude per se, but is likely to benefit from isolation on different peaks, changes in distribution likely related to climate fluctuation, and host repertoire evolution.

Supplementary Material

Data available from the Dryad Digital Repository: https://dx.doi.org/10.5061/dryad.bk3j9kdkz.

Author Contributions

H.W., P.Z., F.M., and M.W. conceived the study. S.M., Y.Z., and M.P.B. collected and analyzed molecular data. M.P.B. and S.N. made host plant data analysis. A.M. and D.J.L. undertook Asteroid and PhyloNetwork analyses. S.M. wrote the original draft with review and edits from D.J.L., M.P.B., S.N., A.M., C.W.W., N.W., P.Z., and H.W.

Conflict Of Interest

The authors declare no conflict of interest.

Funding

This work was supported by grants from (i) the National Natural Science Foundation of China (No. 32070469 to M.W., No. 32270478 to H.W., and No. 32071611 to P.Z.); (ii) the Swedish Research Council (International Postdoc Grant No. 2020-06422 to M.P.B., Grant No. 2019-03441 to S.N.); (iii) the US National Science Foundation (DEB-1541557 to D.J.L.); and (iv) the China Biodiversity Observation Networks for Butterflies (China BON-Butterfly) to F.M.

Data Availability

Code used for software in this study can be also found from the Zenodo using the link: https://doi.org/10.5281/zenodo.13932428