Population genomics reveal multiple independent origins of pesticide resistance in the polyphagous pest, Tetranychus urticae

Abstract

The rapid evolution of pesticide resistance imposes great pressure on food production. However, how resistance alleles arise and spread across field populations remains largely understood. Here, we study the evolutionary trajectories of resistance alleles in Tetranychus urticae, a rapidly evolving pest. We sequence the genomes of 258 T. urticae females collected from China. Combined with global reference genomic data, we examine the evolutionary origin(s) of 18 mutations across 10 target-site genes and analyze the global population genetic structure using genome-wide SNPs. Our findings reveal a striking prevalence of multiple independent origins of resistance mutations, with only two of 18 mutations showing an apparent single origin. Population structure and haplotype analyses point to an important role of gene flow in the spread of resistance alleles. Selection analyses reveal pesticide-driven sweeps affecting genetic diversity. These findings advance our understanding of the rapid adaptation of arthropod herbivores to extreme selective pressure.

Population genomics uncover multiple independent origins and the importance of gene flow in the spread of target-site resistance mutations in the mite pest Tetranychus urticae.

Introduction

The evolution of pesticide resistance in arthropod pest populations offers a clear example of rapid adaptive evolution and provides an excellent model for addressing fundamental questions in the genetics of adaptation, including the roles of new mutations, pre-existing standing genetic variation, and the geographic spread of adaptive alleles through gene flow^1–3.

While the mechanisms underlying resistance have been increasingly deciphered⁴, our understanding of how resistance alleles arise in field populations (from either a single origin or from multiple independent origins) remain incompletely understood and appears to vary among different resistance genes and species/populations^5–13 as reviewed by Hawkins et al.¹⁴. Understanding how resistance alleles evolved in pest populations not only advances our theoretical knowledge of adaptive evolution but also supports practical approaches for integrated pest management^14,15. A comprehensive understanding of resistance evolution requires the study of resistance alleles within the context of population structure^12,16–18. For instance, by comparing structure patterns of the single nucleotide polymorphisms (SNPs) within an resistance gene with those of genome-wide SNPs, evolutionary histories of resistance mutations were successfully traced in Indo-Pacific populations of Aedes aegypti (L.)¹⁶.

The two-spotted spider mite, Tetranychus urticae (Acari: Tetranychidae), is a highly polyphagous agricultural pest, feeding on over 1000 different host plants from more than 140 families, including approximately 150 different crops^4,19. In parallel with its wide host plant range, T. urticae has shown an extraordinary ability to rapidly evolve resistance to pesticide/acaricides, often within just a few years of a new acaricide’s introduction^4,20. Currently, 557 resistance cases involving 96 types of synthetic pesticides have been documented for T. urticae (https://www.pesticideresistance.org/).

Due to the economic importance of T. urticae as well as its utility as a model organism for studying rapid evolution of pesticide resistance, extensive studies have been conducted on the mechanisms of resistance^4,20. The T. urticae genome, initially sequenced using Sanger technology and later assembled at a chromosomal level, provides a high-quality reference for population genomics and genetic mapping studies^21–24. A variety of mechanisms have been elucidated, including increased metabolic detoxification of the pesticide (i.e. toxicokinetic resistance) and by mutations in the target-site of the pesticide (i.e. toxicodynamic resistance). To date, 30 mutations across 10 target-site genes in T. urticae have reliably been linked to resistance to pesticides that are currently categorized to 12 different modes of action. Causality of most of the mutations has been corroborated by robust genetic and biochemical evidence, most recently by gene-editing^25,26. These target-site resistance mutations provide valuable molecular markers to trace the evolutionary trajectories of pesticide resistance.

In China, T. urticae (green morph) is considered as an invasive pest, first detected on Begonia plants in Beijing in 1983, after which it spread widely across China and has become a serious pest on economically important crops²⁷. However, the geographical origin(s) of these invasive populations remains unclear. This pest has also rapidly developed strong resistance to most acaricides following its introduction to China^13,28–31. Despite an expected reduced genetic variation due to potential founder effects, the rapid spread of acaricide resistance in China suggests a complex interplay of gene flow and independent evolutionary events. The mechanisms and relative contributions of single genetic factors to the rapid adaptive evolution of Chinese populations of T. urticae remain poorly understood.

In the current study, we re-sequenced the genomes of 258 T. urticae individuals from 25 distinct populations across China and combined this data with 65 publicly available data-sets of genomic reads from 15 countries^32–39. Here, we aim to decipher the origin(s) of pesticide resistance mutations by haplotype network analyses in conjunction with population genetic structure analyses using genome-wide SNPs. Our population genomic analyses describe the repeatability of target-site resistance evolution, and advance our understanding of rapid adaptation of arthropod herbivores to extreme selective pressure.

Results

Genome re-sequencing and comparison between two sequencing methods

We successfully extracted DNA from 258 adult female individuals collected from 25 field populations across China (Fig. 1a, Supplementary Data 1–2). Subsequently, we individually amplified the entire genome using the REPLI-g Ultrafast Mini WGA kit (QIAGEN) and performed re-sequencing of these whole-genome amplicons on the BGI MGISEQ-2000 platform. Sequencing depth ranged from 13.6 to 74.3 X, with an average depth of 42.1 X. We also integrated 64 re-sequencing reference data from Africa, Asia, Europe, and America (Supplementary Data 3), generated by pool sequencing of hundreds to thousands of female offspring from iso-female or inbred lines. The sequencing depths of these reference genomes ranged from 36.0 to 401.6 X, with an average depth of 78.4 X.

Fig. 1. Sample locations and population structure of global T. urticae populations.

a Geographic sampling locations of T. urticae used in this study. b Rooted neighbor-joining (NJ) phylogenetic tree of 70 representative T. urticae samples based on genome-wide SNP data. Two closely related species T. truncatus and T. pueraricola were used as outgroups. c Rooted NJ consensus tree constructed using 1000 bootstrap replicates. Node labels represent bootstrap values, which indicate the confidence levels of the corresponding branches. T. truncatus and T. pueraricola were included as outgroups. d Principal component analysis (PCA) of representative T. urticae individuals (Supplementary Data 5). e, f PCA analyses of the selected individuals. The NJ tree and PCA analyses were conducted using 304,709 nuclear biallelic SNPs with a sequencing depth of more than 10 X that were present in all samples.

Unlike the reference re-sequencing data from non-Chinese populations, which were directly sequenced from DNA extracted from pooled individuals, our methodology included an additional whole-genome amplification step for the DNA of each single individual. Therefore, we compared the outcomes between the two sequencing methods. We constructed two iso-female lines for each of the two populations reared in laboratory (Supplementary Data 4). The female founders underwent DNA extraction and genome amplification immediately after the populations were established. Their offspring, which were derived from mother-son mating, were expanded for 2 months (~4 generations) to a population size of approximately 400 individuals and then pooled for DNA extraction. We found that the two sequencing methods yielded similar sequencing coverage when calculated at sequencing depths of >1 X and >4 X (Supplementary Fig. 1a, b). After initial filtering, we obtained 2,001,150 SNPs from the 8 samples. The SNP missing rates were slightly higher for the whole-genome amplification samples (mean = 5.98%) compared to the pool sequencing samples (mean = 4.08%, Supplementary Data 4). The heterozygosity rates of the single female samples were slightly higher than those of the offspring pool sequencing samples, both at SNPs with allele depth >4 X (n = 715,484) and >10 X (n = 595,080, Supplementary Data 4). The SNP similarities between each pair of samples from the two sequencing methods ranged from 97.94% to 99.17% (mean = 98.73%) and from 98.46% to 99.66% (mean = 99.26%) at SNPs with allele depth >4 X and >10 X, respectively (Supplementary Fig. 1c, d). In conclusion, the two sequencing methods yielded relatively consistent results, with only slight differences in some indices.

The genetic relationships among worldwide T. urticae

Before investigating resistance evolution, we initially characterized the global population structure of T. urticae. As the reference whole-genome re-sequencing samples mostly originate from one individual per sampling location, we avoided potential bias by initially selecting only one individual with the highest sequencing depth from each population for these global population genetic structure analyses (Supplementary Data 5). The Neighbor-Joining (NJ) tree based on the total 304,709 nuclear SNPs across 70 samples showed that the European samples occupied a basal position in the tree and exhibited high genetic divergence when rooted with T. truncatus and T. pueraricola, suggesting that T. urticae might have originated within Europe (Fig. 1b, c). The Asian samples generally formed a monophyletic lineage, occupying a terminal position in the tree with robust support, suggesting a more recent divergence from other populations (Fig. 1b, c). Samples from the American continent (Canada, USA, Brazil) and Africa (Ethiopia) were interspersed within the European lineages roughly at the transition between the Asian and European lineages in the NJ tree.

Similar results were obtained by a principal component analysis (PCA) based on the 304,709 nuclear SNPs across 70 samples. However, it showed that the Romanian sample, which was located at the basal clade in the NJ tree, was quite distinct from the others (Fig. 1d). This sample was found to possess a considerable proportion of the T. turkestani genome in a previous study, likely due to interspecific hybridization³⁶. The other samples were clustered in a line, in which the non-European samples and six samples from Italy, Greece, France, and England were tightly clustered at the lower part of the panel (Fig. 1d, e). Although tentative, these clustering patterns further pointed to Europe as the origin of T. urticae. In addition, PCA analysis using a subset of samples clustered together revealed that the Chinese samples had a much closer relationship with the samples from Greece, France, England, and Ethiopia, suggesting these areas may be a source of the Chinese invasive populations (Fig. 1f).

We also constructed a mitochondrial SNP network by encompassing all samples (n = 322, Supplementary Data 2, 3) using 1827 SNPs (Fig. 2). In line with the nuclear findings, the mitochondrial network revealed that European lines possess highly divergent haplotypes. In contrast, Chinese samples exhibited extremely similar haplotypes, and most of which were shared by the samples close to Chinese samples on the nuclear SNPs. Within the Chinese samples, we found two low-divergence haplogroups (HGI and HGII). The two haplogroups differ by 11 SNPs, including two non-synonymous substitutions, which are located in COXIII and ND4.

Fig. 2. A global mitochondrial haplotype network for all T. urticae samples (n = 322).

Haplotype network was constructed using 1827 mitochondrial SNPs. The pie size is proportional to the number of individuals sharing each haplotype, and the colors represent their geographic distribution. The values in parentheses and dashes between haplotypes indicate the number of mutational steps. The haplotypes of Asian individuals, along with adjacent haplotypes, were enlarged and categorized into two groups, HGI and HGII. The two haplogroups differ by 11 SNPs, including two non-synonymous substitutions, which are located in COXIII and ND4.

Compared to the European lines as a group, the Chinese individuals collectively displayed a serious decrease in the number of SNPs (Supplementary Fig. 2a, b), particularly for SNPs with low minor allele frequency (MAF) (e.g., approximately 50% decrease for SNPs with MAF < 0.2, Supplementary Fig. 2b). The level of heterozygosity in Chinese populations was lower than in European populations (P = 0.0014, Supplementary Fig. 2c). This diversity pattern indicates that the Chinese populations have experienced a strong founder effect.

The Chinese populations can be categorized into two genetic groups with gene flow

To further characterize the structure of Chinese populations, we generated a dataset comprising 258 individuals exclusively from China. This dataset consisted of 19,480 SNPs that were unlinked (r² < 0.2) and present in all Chinese individuals. Population pairwise F_ST of the Chinese populations ranged from −0.009 to 0.407, with a mean of 0.167 (Supplementary Fig. 3a). A Mantel test detected a significant correlation between genetic and geographic distances (R = 0.330, P < 0.001; Supplementary Fig. 3b), indicating a role of geographical isolation in shaping population structure.

The PCA analysis clearly delineated Chinese individuals into three main groups with some association to geographic distribution (Fig. 3a). Group I was consisted of five Southwestern populations (YNkBe, YNkEg, YNkRo, YNkSt, and GZqMa) and one individual from JSnSt, which separated from others along PC1 (explaining 10.36% variance). The remaining samples were further separated into two groups (II and III) along PC2 (explaining 8.87% variance), with individuals from the XJeBe population forming Group III. To quantify the genetic differentiation between these groups, we calculated F_ST and Dxy (the absolute nucleotide divergence) between the three groups. The F_ST values between Group I and II were much lower (0.096) than the other two comparisons (Group I vs. Group III: 0.288; Group II vs. Group III: 0.222; Fig. 3a). However, Dxy values were comparable among pairwise comparisons of the three groups, ranging from 0.164 to 0.188, with slightly higher values observed between Group I and II (Fig. 3a). This discrepancy is likely attributed to within-population diversity, as F_ST is more sensitive to such variations than Dxy⁴⁰. Reduced within-population diversity observed in the XJeBe population (Supplementary Fig. 4) likely contributed to the higher F_ST values between Group III and the other two groups.

Fig. 3. Population structure of the Chinese populations of T. urticae.

a PCA plot of all Chinese individuals (258 individuals from 25 populations in Supplementary Data 1) based on nuclear SNPs. b Relationships and gene flow events among Chinese populations inferred by the maximum-likelihood method implemented in TreeMix. The ‘Nuclear PCA’ column indicates the group, as defined in the PCA plot (Fig. 3a), to which each population was assigned. The ‘Mitochondrial haplogroup’ column indicates the mitochondrial haplogroup, as defined in the mitochondrial haplotype network (Fig. 2), found in each population. c Admixture analysis of genetic structure and individual ancestry. The optimal number of ancestral clusters chosen based on cross-validation error was 15. Each individual is represented by a vertical bar displaying membership coefficients to each of the 15 genetic clusters in (c). The above three analyses were all based on 19,480 SNPs that were unlinked (r² < 0.2) and present in all Chinese individuals.

We explored these relationships further using TreeMix, which builds a maximum-likelihood (ML) tree from population allele frequencies. We ran TreeMix on the 25 Chinese populations with a T. truncatus population (Supplementary Data 6) used as the outgroup. The ML tree showed that the Chinese populations could be grouped into two clades (Fig. 3b). These clades generally corresponded to Group I and II observed in the PCA plot (Fig. 3a), although exceptions were noted for the YNkSt and XJeBe populations. Specifically, the YNkSt population was placed within Group I in the PCA plot but within Clade II in the ML tree. Similarly, the XJeBe population formed a distinct Group III in the PCA plot but clustered within Clade II in the ML tree. Treemix also identified gene flow from Clade I (JSnSt and YNkSt) to Clade II (YNkBe) and between populations within Clade I. In support of the existence of two genetic groups in our Chinese populations, two mitochondrial haplogroups were observed (Fig. 2). However, a portion of individuals with the nuclear background of Group II possessed mitochondrial haplotypes belonging to the Group I mitochondrial haplogroup, suggesting hybridization between the two clades may have occurred. In addition, an apparent introgression event was also detected from T. truncatus into the HAzGr population of T. urticae.

ADMIXTURE analysis partitioned genetic variation into 15 genetic groups (i.e., optimal K = 15, Supplementary Fig. 5), where samples were clustered according to their collection sites (Fig. 3c). This relatively strong population structure might reflect elevated levels of inbreeding or relatedness within populations. Admixture of genetic ancestries was also observed in certain populations: JLcSt, JLcPe, QHhSt, HAzGr, JSnSt, NMhSt, and GZqSt shared similar coefficients in inferred genetic ancestry, indicating some degree of inter-population gene flow, which aligns with those gene flow events identified by TreeMix. Notably, both ADMIXTURE and TreeMix suggested significant gene flow among four populations collected from strawberry (JLcSt, QHhSt, JSnSt, and GZqSt), despite distances of hundreds to thousands of kilometers between the sampling locations, suggesting potential anthropogenic factors such as stolon transfer influencing T. urticae dispersal in China.

Collectively, the Chinese populations exhibit moderate differentiation with substantial gene flow among them, particularly noticeable among populations collected from strawberries. It appears that Chinese populations consist of two genetic groups with similar levels of genetic diversity. These two groups exhibit marked differences in nuclear and mitochondrial variation, implying two potential invasion origins. A smaller group comprising four populations (Group I in the PCA plot) is confined to southwestern China, whereas the larger group (Groups II + III) is more widely distributed across China. The XJeBe population (constituting Group III) likely originated from a single colonization event, descending from a subpopulation within Group II.

Spread of target-site resistance mutations in Chinese and European populations

In the Chinese populations, 18 out of 30 target-site mutations were detected in 10 genes associated with pesticide resistance (Fig. 4). We adopted the common nomenclature for numbering resistant mutations in acetylcholinesterase (AChE), voltage-gated sodium channel (VGSC) and the PSST homologue of complex I (PSST), based on the sequences of Torpedo californica, Musca domestica, and Yarrowia lipolytica respectively. The corresponding numbering information in T. urticae is also accessible in Supplementary Data 7. The mutation F331W/Y/C in AChE, known to confer resistance to carbamates and organophosphates, was the most prevalent, occurring in 95.3% of the Chinese samples. This prevalence likely reflects extensive historical use of these compounds since the 1940s^36,41. In contrast, the frequency of F331W/Y/C was lower in the European populations (65.6%), likely due to the more limited use of organophosphates in Europe in the last 10 years.

Fig. 4. Occurrence of mutations associated with pesticide resistance in ten target genes in Chinese and European populations of T. urticae.

‘GT’ represents genotype, ‘R/R’ represents resistance mutation homozygote, ‘R/S’ represents resistance mutation heterozygote, S/S, susceptible homozygote. SdhB dehydrogenase B, SdhC dehydrogenase C, CHS1 chitin synthase 1, ATPs ATP synthase subunit C, PSST PSST homologue of complex I, GluCl1 glutamate-gated chloride channel subunit 1, GluCl3, glutamate-gated chloride channel subunit 3, AChE acetylcholinesterase; VGSC voltage-gated sodium channel, Cytb, mitochondrial cytochrome b. The amino acid substitutions in AChE, VGSC and PSST follow the numbering conventions of commonly used reference species, while substitutions in the other genes follow the T. urticae numbering. For the Chinese samples, 258 individuals from 25 field populations were surveyed. The two genetic groups (Group I and II) are defined according to the population genetic structure results based on nuclear and mitochondrial variation (Fig. 3). European samples included 19 T. urticae lines from 9 countries.

The second most frequent target-site resistance mutation in our Chinese collection of populations was A1215D in VGSC (92.4%), which alone does not confer resistance to pyrethroids but always appears with F1538I⁴². F1538I can confer high resistance and it occurs at a frequency of 69.1%. In contrast, L1024V in VGSC, which alone can cause exceptionally strong resistance to pyrethroids (bifenthrin, fenpropathrin, and fluvalinate)⁴² was at low frequencies (2.5%) and always in heterozygous state at the individual level. The well-established knockdown resistance (kdr) mutation L1014H was absence in our samples. In addition, the super-kdr mutation M918L/T described in insects and the spider mite T. evansi⁴³ was also not found.

A higher frequency was also detected at H92R in PSST (78.9%), which confers resistance to the acaricides of mitochondrial complex I inhibitors (METI-I acaricides)⁴⁴, and is also widely distributed in Europe⁴⁵. Similarly, G314D and G326E in glutamate-gated chloride channel subunit 1 (GluCl1) and glutamate-gated chloride channel subunit 3 (GluCl3), respectively, also occurred at high frequencies (74.0% and 75.9%), which are major factors contributing to abamectin and diphenylcarbinol acaricide resistance by a synergistic manner in T. urticae⁴⁶. Notably, these two mutations were also strongly linked (D’ = 0.844) in the Chinese individuals despite being over 16 Mb apart on Chromosome 2. Relatively lower frequencies were observed at H146Q (11.7%), I260T/V (21.2%) in succinate dehydrogenase B (SdhB), and at S56L (0.2%) in succinate dehydrogenase C (SdhC), which confer resistance to newly developed mitochondrial complex II inhibitors (METI-II acaricides). No resistance mutation in mitochondrial cytochrome b (Cytb) was found in the Chinese populations.

We found different resistance profiles between the two genetic groups we defined, with populations in Group I exhibiting much lower frequencies of resistance mutations compared to those in Group II (Fig. 4). Interestingly, I321T in GluC13, which contributes to abamectin resistance and mainly occurs in the red morph of T. urticae^11,36,47, was only detected in Group I. This mutation was also previously detected in samples from France and Ethiopia³⁶. This result provides additional support for the genetic divergence between the two genetic groups.

Evolutionary origins of the target-site resistance mutations in the Chinese T. urticae

To reveal the evolutionary modes of the resistance mutations, we phased the SNPs within the 8 nuclear resistance genes into haplotypes and constructed a haplotype network for each resistance allele using all available global samples (n = 322). Notably, SdhC was excluded from this analysis because only a single resistant heterozygote was detected for this gene. Haplotype networks showed that only G314D in GluCl1 and V89A in ATPs occurred on the same or near-identical haplotypes (Fig. 5a and Supplementary Fig. 6a), strongly pointing to a single origin. The H146Q in SdhB arose dominantly from a group of closely related haplotypes and also to some extent from distantly related haplotypes (Supplementary Fig. 7a), likely resulting from recombination between haplotypes. The rest of the 15 resistance mutations arose from multiple origins as they occurred on extremely distinct haplotypes (Fig. 6a and Supplementary Fig. 8a–10a). This pattern persisted when analyzing the European and invasive Chinese populations separately.

Fig. 5. Network of the GluCl1 haplotypes and geographic distribution of major haplotypes carrying target-site resistance mutations.

a Haplotype network for global samples (n = 322, 40 populations). Each pie represents a unique haplotype, with its size being proportional to the number of individuals sharing that haplotype. Dashes in the connections between haplotypes represent mutational steps. Red pies represent resistant haplotypes, which carry the G314D resistance mutation. Grey pies represent susceptible haplotypes, which do not carry the G314D resistance mutation. b Occurrence of major resistant haplotypes in different countries (Supplementary Data 8). Hap_3 and Hap_19 are the two major resistant haplotypes. The colors of pie chart represent the proportion of haplotypes from different countries. GR Greece, IT Italy, CN China. c Geographic distribution of major resistant haplotypes in Chinese populations (n = 258, 25 populations as listed in Supplementary Data 1). “S” represents haplotypes carrying the susceptible allele. “Others” refers to other resistant haplotypes occurring at low frequencies. “Hap_19” denotes the predominant haplotype carrying the G314D resistance mutation.

Fig. 6. Network of the GluCl3 haplotypes and geographic distribution of major haplotypes carrying target-site resistance mutations.

a Haplotype network for global samples (n = 322, 40 populations). Each pie represents a unique haplotype, with its size being proportional to the number of individuals sharing that haplotype. Dashes in the connections between haplotypes represent mutational steps. The colors of the pie charts represent different genetic profiles, categorized based on the alleles at the two target sites, I321T and G326E, in GluCl3. The two letters in “RS”, “SR”, and “SS” correspond to I321T and G326E, with “R” and “S” representing resistance and susceptible alleles respectively. b Occurrence of three predominant resistant haplotypes across different countries (Supplementary Data 8). Hap_44, Hap_46 and Hap_145 are three predominant haplotypes carrying the G326E resistance mutation. The colors of pie chart represent occurrence in different countries. IT Italy, UK United Kingdom, CN China. c Geographic distribution of predominant resistant haplotypes in Chinese populations (n = 258, 25 populations as listed in Supplementary Data 1). “SS” represents haplotypes carrying susceptible alleles at both target sites, and “Others” represents other resistant haplotypes occurring at low frequencies.

An interesting pattern was observed for A1215D and F1538I in VGSC. Each of the two mutations occurred on distantly related haplotypes, but their combination mainly occurred in a group of closely related haplotypes (Fig. 7a). This pattern indicates the two mutations arose independently and from multiple origins. A1215D alone does not confer resistance to pyrethroids⁴² but F1538I can^48,49. The two mutations simultaneously occurred on multiple distantly related haplotypes but were most frequent in a group of closely related haplotypes (Hap_36 related, Fig. 7a, c), which were widely shared by different populations in China. The occurrence of major resistance haplotypes is listed in Supplementary Data 8.

Fig. 7. Network of the VGSC haplotypes and geographic distribution of major haplotypes carrying target-site resistance mutations.

a Haplotype network for global samples (n = 322, 40 populations). Each pie chart represents a unique haplotype, with its size being proportional to the number of individuals sharing that haplotype. Dashes in the connections between haplotypes represent mutational steps. The colors of the pie charts represent different genetic profiles, categorized based on the alleles at the three target sites, L1024V, A1215D, and F1538I. The three letters in “RRR”, “RRS”, and “RSS” etc. correspond to L1024V, A1215D, and F1538I, with “R” and “S” representing resistance and susceptible alleles, respectively. b Occurrence of two major resistant haplotypes across different countries (Supplementary Data 8). Hap_3 and Hap_36 are two major haplotypes carrying target-site resistance mutations. CA Canada, GR Greece, BE Belgium, DE Germany, CN China. c Geographic distribution of predominant resistant haplotypes in Chinese populations (n = 258, 25 populations as listed in Supplementary Data 1). “SSS” represents haplotypes carrying susceptible alleles at the three target sites, and “Others” represents other resistant haplotypes occurring at low frequencies.

To validate the haplotype network results, we conducted PCA analyses using SNPs within the target-site genes but excluding the target-site resistance mutation sites. Our reasoning was that if a resistance mutation originated from a single source and spread through gene flow, then the resistant homozygotes would cluster together in the PCA plots based on SNPs within the target-site gene regardless of the field population from which they were collected. In support of the haplotype network results, the PCA analyses also revealed that only the homozygotes for the mutations G314D in GluCl1, V89A in ATPs, and H146Q in SdhB were closely clustered together (Supplementary Fig. 11a, c and e). In addition, the homozygotes of the A1215D + F1538I combination in VGSC occurred at a high frequency and largely formed a single main cluster, though some were spread apart (Supplementary Fig. 11h).

Widespread gene flow among T. urticae populations promotes the spread of resistance

Certain resistant haplotypes of GluCl1, GluCl3, VGSC, ATPs, PSST, and CHS1 were shared by individuals from different countries (Figs. 5b−7b, Supplementary Figs. 6b, 8b and 9b), reflecting a global spread of resistance mutations by gene flow. Of note, the Chinese populations shared resistant haplotypes with the European populations in GluCl3, VGSC, ATPs, PSST, and CHS1, suggesting invasions of these resistance alleles. Geographic distributions of the major resistant haplotypes in the Chinese populations showed strong resistance gene flow among populations, regardless of whether the mutations arose from a single or multiple independent origins (Figs. 5c−7c, Supplementary Figs. 6c−9c). We noticed that a major haplotype carrying H146Q in SdhB (Hap_26) was largely associated with the populations collected from strawberries (JLcSt, QHhSt, JSnSt, NMhSt, and GZqSt, Supplementary Fig. 7c). This agreed with the strong gene flow among strawberry populations as inferred using genome-wide SNPs, further pointing to a potential anthropogenic factor in spreading of resistance mutations.

Selection pressure on the Chinese populations of T. urticae

To explore whether T. urticae has undergone adaptive evolution after its invasion into China, we employed the Composite Likelihood Ratio (CLR) statistic for genome-wide selection analysis⁵⁰. The CLR test compares the likelihood of observed genetic data under a model of selection to that under a neutral model, effectively identifying genomic regions that have undergone positive selection. This method is particularly useful for detecting selective sweeps, where advantageous alleles rapidly increase in frequency, leaving distinct signatures in the genome. We used the field-collected Chinese samples (n = 258) for the selection analyses.

We considered genomic regions with the top 1‰ CLR values to be under significant selective sweeps. Above this threshold, 9 peaks (P1-P9) were identified (Fig. 8, Supplementary Data 9–17). The most prominent peak (P5) was located at 25.00−25.79 Mb on chromosome 2, containing 196 genes, of which 11 are gustatory receptor genes (Fig. 8b, Supplementary Data 13). Recent studies have found that chemosensory protein receptor genes can be associated with resistance and host plant acceptance and exploitation^24,51,52. This raised the possibility that selection pressures could be associated with pesticide use. To test this possibility, we superimposed the 9 nuclear resistance-related genes onto the CLR results. We found two selection signal peaks (P6 and P9) strongly associated with VGSC and GluCl1 (Fig. 8), in which resistance mutations or haplotypes were inferred to be single origin and with high frequencies. The regions near the two genes showed reduced nucleotide diversity and notably low negative Tajima’s D values, indicative of hard sweeps. To validate that the detected selection signals are indeed associated with resistance, we separately selected homozygous resistant and susceptible individuals, based on the G314D mutation in GluCl1. We performed CLR analyses and calculated nucleotide diversity (PI) across the genome for each group. The results showed that the signal near GluCl1 was completely absent in the susceptible individuals, while it became stronger in the resistant individuals (Supplementary Fig. 12). This further supports the association between the observed signals and resistance. Additionally, we found significantly lower PI values in the resistant individuals compared to the susceptible ones in the GluCl1 region (Supplementary Fig. 12), which further supports the single origin of the G314D mutation.

Fig. 8. Genomic selective sweep scan for Chinese populations of T. urticae.

a panels display CLR values, nucleotide diversity (PI), and Tajima’s D values across the 3 chromosomes of T. urticae individuals from China (n = 258, 25 populations listed in Supplementary Data 1). Blue points in the CLR panel denote regions that overlap with known pesticide resistance-related genes or QTLs. Tajima’s D, and nucleotide diversity (PI) were estimated across the entire genome using a sliding window approach, with a window size of 50 kb and a step size of 5 kb. Blue dashed lines mark the peaks of CLR signals, where adjacent regions show significantly reduced nucleotide diversity and Tajima’s D values. The dashed red line shows the threshold of significant CLR values (1‰ CLR values). b Genes within the most prominent peak (P5). Vertical dotted lines represent the midpoint of the top genomic window. TuGR: gustatory chemosensory receptor.

We also projected 9 quantitative trait locus (QTL), which were previously demonstrated to be associated with resistance by BSA analyses, onto the CLR result (Supplementary Data 18, Fig. 8). We found that P1 and P3 coincided with a pyrethroid (bifenthrin)-resistance QTL and a shared QTL associated with amitraz and chlorfenapyr resistance, respectively. The pyrethroid (bifenthrin)-resistance QTL located on Chromosome 1 has a broad range (16.1–17.4 Mb) and spans a region of more than 250 genes, including 3 Glutathione S-transferases and 4 ABC transporters previously reported to be involved in the detoxification process³⁹. The shared QTL associated with amitraz and chlorfenapyr resistance was located at 20.85–21.08 Mb on Chromosome 1, which involves 13 cytochrome P450 genes of the CYP392A family, a carboxyl/cholinesterase (CCE20, tetur03g00310), a short-chain reductase (tetur03g00300), and a cluster of four nuclear hormone receptors⁵³. In addition, a second amitraz-resistance QTL and an abamectin-resistance QTL also exhibited signals of selective sweep. The second amitraz-resistance QTL was located at ∼3.008 Mb on Chromosome 3, in which multiple genes coding for cytochrome P450 enzymes of the CYP392A family were present, with CYP392A10v2 (tetur02g14400) being the most centrally located in the peak⁵³. The abamectin-resistance QTL was located at ~6.6 Mb on Chromosome 3, which involved two genes encoding degenerin/Epithelial Na + channels (ENaCs), 19 genes encoding chemosensory receptors, and a gene encoding an inositol monophosphatase-like enzyme (tetur02g06900)⁵⁴.

In the remaining selective peaks, detoxification, excretion and chemoreception-related genes were also found, including carboxyl/cholinesterase (at P2, P8; Supplementary Data 10, 16), cytochrome P450 (at P2; Supplementary Data 10), ABC-transporter (at P2; Supplementary Data 10), solute carriers (at P4; Supplementary Data 12), major facilitator superfamily (at P2; Supplementary Data 10), intradiol ring-cleavage dioxygenase (at P8; Supplementary Data 16), gustatory receptor (at P7, P8; Supplementary Data 15, 16). Notably, these genes are not clearly associated with pesticide use.

Discussion

Pesticide resistance remains one of the most pressing issues in sustainable crop protection and food security, and the mechanisms underpinning its evolution are a major focus of adaptation research. Population genomics has helped identify how different genetic resistance profiles are produced and maintained. In this study, we used a large population genomic dataset to reveal the global population genetic structure of T. urticae, with a particular focus on the Chinese populations. Building upon understanding of population genetic structure, we investigate how the target-site resistance mutations evolved both globally and in China.

Prior to investigating the evolution of target-site resistance, we first characterized the population structure of T. urticae. Although tentative, our results suggest that Europe might be the cradle of T. urticae. This hypothesis is consistent with the basal position of the European populations in the NJ tree and their higher genetic diversity in both nuclear and mitochondrial genomes. Furthermore, in contrast to the non-European samples, which formed a compact cluster, the European samples were widely dispersed in the PCA plots, indicating a high degree of divergence among them. This could reflect a much longer evolutionary history of T. urticae within Europe. However, due to the scarcity of global samples outside of Europe and Asia, this hypothesis awaits further formal testing with additional samples from other geographic regions. In comparison to the European populations, the Chinese populations displayed lower genetic diversity. Additionally, the more pronounced decline in allelic diversity compared to heterozygosity suggests that the Chinese populations have undergone strong founder effects following their introduction.

The Chinese samples clustered with a group of individuals from four other continents in the PCA plots (Fig. 1), suggesting low divergence or on-going gene flow among continents. This makes it difficult to pinpoint the exact invasion origin(s) of the Chinese populations. Allele frequency-based analyses with more global individuals are necessary to reliably trace the origin of the Chinese population. In line with this population structure, some resistance haplotypes of GluCl1, GluCl3, ATPs, SdhC, Cytb, and VGSC were widely shared by individuals from different countries, suggesting that global gene flow is contributing to the development and spread of pesticide resistance.

Both the nuclear and mitochondrial variation revealed the presence of two moderately divergent genetic groups within the Chinese populations, probably reflecting two invasion origins of the Chinese T. urticae. These two groups had similar genetic diversity but obviously different resistance mutation profiles, with Group II having much higher frequencies of target-site resistance mutations compared to the Group I. A higher frequency of resistance mutations might explain why Group II had a much broader geographic distribution than Group I, which appeared to be restricted to Southwestern China. This pattern also raised the possibility that the resistance mutations were carried into China by the invaders.

In addition to frequencies, the two genetic groups also differ in the mutation types. The I321T in GluC13, which alone contributes to abamectin resistance^11,36,47, was detected exclusively in three populations belonging to Group I. A previous study showed that this mutation mainly occurred in the red morph of T. urticae. Only one instance was found in 18 female lines of the green morph of T. urticae, which has been demonstrated to be due to introgression from the red morph³⁶. However, the haplotypes carrying I321T in the Chinese populations were highly divergent from the European populations, suggesting multiple independent origins.

Gene flow was detected between Chinese populations, from JSnSt and YNKSt, which belong to Group II and I respectively, to YNkBe, which belong Group I (Fig. 3b). Six out of 16 individuals from YNkBe carried a Group I nuclear background but a Group II mitochondria, reflecting a higher level of mitochondrial gene flow than nuclear. This pattern agrees with a model proposed by Patten and coauthors⁵⁵, in which introgression rates are predicted to be higher for mitochondrial DNA than for nuclear DNA in haplodiploid taxa.

Pairwise F_ST values (−0.009 to 0.407) revealed moderately high genetic differentiation among Chinese T. urticae populations, which were notably lower than those observed in two Chinese native species (T. truncatus, 0.041–0.809; T. pueraricola, 0.115–0.748)⁵⁶, which are the most dominant spider mites in field crops across China^57,58. Therefore, we suspect that transport of greenhouse plants has largely facilitated the spread of T. urticae in China. Genetic similarity and admixture among five strawberry-collected populations (JLcSt, QHhSt, JSnSt, NMhSt, and GZqSt,) that are separated by hundreds to thousands of kilometers provide strong evidence for this hypothesis.

In total, we detected 18 out of 30 target-site mutations in 10 genes in the Chinese populations, with higher frequencies of target-site resistance mutations related to long-time used acaricides (e.g. carbamates, organophosphates, pyrethroids, METI-I acaricides, and abamectin) than those conferring resistance to newly developed acaricides (e.g., METI-II acaricides). These frequencies are similar to those reported in a recent study by Zong et al.⁵⁹, which employed an amplicon sequencing method. Overall, resistance mutation frequencies in Chinese populations were consistently higher than those in European populations, except for mitochondrial mutations linked to bifenazate resistance^4,60,61, which were absent in Chinese populations. This finding aligns with the results of Zong et al.⁵⁹. However, Zong et al. also noted the absence of mutations in populations with high resistance to bifenazate, suggesting the presence of resistance mechanisms other than target-site based in Chinese populations, such as metabolism resistance mediated by P450⁶². The exact reasons for the higher frequency of resistance mutations in Chinese populations compared to European populations remain unclear, but differences in pesticide usage frequency and varying resistance of T. urticae across different crops sampled may be important contributing factors.

Of the 18 target-site mutations detected, only two mutations fit a single origin pattern. Multiple independent origins involving diverse resistance-related mutations have been frequently reported in other insects^10,63–66, weeds^1,67,68, and pathogens^69,70 (also see the review by Hawkins et al.¹⁴). Among the 21 resistance cases reviewed by Hawkins et al.¹⁴, 9 cases clearly fit a multiple origin pattern, and only 2 cases fit a single origin pattern. The prevalence of multiple independent origins of the resistance mutations reflects the high repeatability of resistance evolution. Different sites within the same gene can exhibit distinct patterns of origin. For example, H146Q in SdhB arose from a single origin, whereas I260T/V arose from multiple origins. The two mutations both confer resistance to acaricides of Complex II inhibitor class. H146Q confers high levels of resistance to cyflumetofen and pyflubumide, but not cyenopyrafen or cyetpyrafen²⁶, whereas I260T/V confers resistance towards cyflmetofen and cyenopyrafen but not pyfblumide⁷¹.

Here, we provide evidence that shows that recombination likely contributed substantially to resistant evolution by bringing together different substitutions in VGSC. A similar phenomenon has been reported in the resistance to fenvalerate in Helicoverpa armigera, where a unique P450 enzyme (CYP337B3) arose from unequal crossing-over between two parental P450 genes and is capable of metabolizing fenvalerate into 4ʹ-hydroxyfenvalerate⁷². Our genomic data suggests that VGSC haplotypes carrying both A1215D and F1694 likely originated from a single recombination event. Previous studies have demonstrated that F1538I alone can confer high resistance to pyrethroids^48,49, whereas A1215D does not confer resistance on its own⁴². The close proximity of these two mutations makes it difficult to uncouple their effects⁴, raising the question of whether A1215D has an additive or synergistic effect on resistance levels. Alternatively, A1215D may compensate for the fitness cost associated with L1024V and F1538I mutations, as suggested by the significantly higher frequencies of haplotypes carrying L1024V + A1215D (2.0%) or A1215D + F1538I (53.0%) compared to haplotypes with L1024V (0.2%) or F1538I (3.3%) alone. A previous study found that the L1014F mutation reduces the heat tolerance in Drosophila melanogaster⁷³. However, fitness costs in T. urticae were not apparent in a study with near-isogenic lines under laboratory conditions (25 ± 1 °C, 60% relative humidity)⁷⁴. Nevertheless, it remains unclear whether the L1024V and F1538I mutations in T. urticae incur a fitness cost at high temperatures, or whether the A1215D mutation can compensate for the fitness cost associated with these two mutations. Further studies are needed to explore the interactions between A1215D and the two resistance mutations.

Shared haplotypes among different populations highlight the crucial role of gene flow in the spread of resistance mutations among different countries. Moreover, apparent gene flow between T. truncatus and T. urticae (Fig. 3b) raises the possibility of interspecific gene flow of resistance alleles in China, particularly for closely related species, such as T. pueraricola and T. kanzawai.

We found genomic regions exhibiting signatures of selection were most likely associated with acaricide resistance in the Chinese populations. In support of this possibility, strong selection signals in the Chinese populations were identified at the GluCl1 and VGSC loci. Particularly for GluCl1, the selection signal near it was completely absent in individuals bearing susceptible genotype, while it became stronger in the individuals with resistant genotype (Supplementary Fig. 12), providing strong support for the association between the observed signals and resistance.

Interestingly, we found the strongest selection signal at a genomic region containing 10 gustatory receptor genes (GRs). Ingham et al.⁵¹ found that a sensory appendage protein (SAP2) is involved in conferring pyrethroid resistance in Anopheles gambiae. They also found that a selective sweep had occurred near the genomic region containing SAP2⁵¹. Since then, the contribution of chemosensory proteins in resistance has received considerable attention^75–78. However, there have been no reports of sequestration resistance mediated by GRs. A more recent study in T. urticae^24,51, using genetic mapping, found that GRs and degenerin/epithelial Na+ channel chemosensory receptors (TuENaC) were enriched in a QTL associated with adaptation to tomato, a challenging host for T. urticae, implying that GRs or TuENaC are involved in resistance to toxic compounds.

Our results suggest that the detectability of selection signals largely depends on the origin patterns and frequencies of the resistance mutations. Only target-site resistance mutations with a single origin and high occurrence frequencies exhibit strong signal. For instance, although G314D (74.0%) in GluCl1 and G326E (75.9%) in GluCl3 had similar high mutation frequencies in the Chinese populations, a selection signal was not observed on GluCl3, where G326E arose from multiple origins. This discrepancy arises because multiple origins preserve greater variation around the resistance mutation due to different haplotype backgrounds¹⁴. The extensive region surrounding GluCl1 under selective sweep strongly suggests that the mutation is relatively recent, likely representing a de novo mutation.

In conclusion, we revealed important features of the global population genetic structure of T. urticae using genome-wide SNPs. Our findings tentatively suggest that T. urticae originates from Europe, with significant gene flow occurring among continents that has facilitated the global spread of target-site resistant mutations. Chinese populations consist of two moderately divergent genetic groups with distinct resistance profiles, likely resulting from two separate invasion events. Target-site resistance mutations in T. urticae predominantly arose from multiple independent origins, indicating a high repeatability of resistance evolution. Pesticide applications have exerted strong selection pressure on genomic regions where resistance mutations emerged from a single origin and occurred at higher frequencies. This underscores the necessity of studying the consequences of pesticide application on pest behavior in future research.

Methods

Spider mite materials and genome re-sequencing

A total of 258 spider mite females (green morph) were collected from 18 cities across China (Fig. 1a and Supplementary Data 1) and identified using the nuclear ribosomal internal transcribed spacer (ITS) region⁷⁹. High-quality DNA was individually extracted from each female using DNeasy Blood & Tissue Kits (QIAGEN, Germany) and then amplified at 30 °C for 16 h using REPLI-g Ultrafast Mini WGA kit (QIAGEN) to produce enough DNA for high-throughput sequencing. The library with almost 350 bp insert size was constructed with MGIEasy FS DNA Library Prep Set (MGI, Shenzhen) and sequenced with a mean genome depth of 42.1 X (ranging from 13.6 to 74.3 X) using the MGISEQ-2000 platform. We also sequenced 7 Tetranychus truncatus females sampled from cotton as outgroup for the following gene flow analyses.

To explore the global population evolution of T. urticae, we also used 64 genome re-sequencing data of 64 T. urticae lines from Africa, Asia, Europe, and America^32–38 (Supplementary Data 3). Most of these genomes were sequenced from iso-female lines or inbred lines, whose DNA were extracted from a pool of hundreds of females for each line to obtain sufficient DNA for sequencing without amplification.

Because the sequencing methods used were different between the non-Chinese samples and our samples, we evaluated the sequencing consistency between them. Briefly, we selected two populations maintained in the laboratory, and from each population, two virgin females were chosen to establish an iso-female line, resulting in a total of 8 iso-female lines (Supplementary Data 4). An iso-female line was created by mating a single virgin female to her first son³⁷. The female founder underwent DNA extraction and subsequent genome amplification using the REPLI-g Ultrafast Mini WGA kit (QIAGEN) after the populations were established. And its offspring were expanded for 2 months (~4 generations) to a population size of ~400, then was pooled for DNA extraction. Next, we performed re-sequencing on the BGI MGISEQ-2000 platform. Genome alignment and variant calling were performed for these samples separately following the next method section. We calculated the whole sequencing coverage at the sequencing depth of 1 X and 4 X using bamdst v1.0.9 (https://github.com/shiquan/bamdst). After SNP calling, we calculated the SNP missing rate and heterozygosity for each sample using PLINK v1.90b4.6⁸⁰. The genome-wide SNP similarity between the single female founder (which underwent whole-genome amplification) and its offspring pool-sequencing data for each iso-female line was also calculated by PLINK v1.90b4.6⁸⁰.

Genome alignment and variant calling

The low-quality reads, which had adapters, an average base quality of both ends below 20, as well as reads with 5 more Ns and length shorter than 15 bp after trimming, were removed from the pair-end 150 bp raw data using BBtools 38.42⁸¹. Then, the clean reads were aligned to the pseudo-chromosomal reference of T. urticae²¹ with the aligner BWA v0.7.17⁸², followed by sorting and indexing with SAMtools 1.13⁸³. Duplicated reads were marked and mapping quality tags were added with Picard tools v2.26.6 (http://broadinstitute.github.io/picard). Variant calling was performed with HaplotypeCaller and GenotypeGVCFs programs in GATK v4.2.2.0⁸⁴. We filtered SNPs based on the distribution of variant annotations, removing those with QUAL (Phred scaled probability) < 30.0, QD (variant confidence standardized by depth) < 2.0, MQ (mapping quality of SNP) < 30.0, FS (strand bias in support for REF versus ALT allele calls) > 10.0, and SOR (sequencing bias in which one DNA strand is favored over the other) > 5.0, MQRankSum (Mapping Quality Rank Sum Test) < −5.5, ReadPosRankSum (Read Position Rank Sum Test) < −3.0 (Set1). Furthermore, SNPs within 10 bp of indels, multiallelic SNPs, sites with missing rates of more than 10% and minor allele frequency (MAF) less than 0.01 were removed using BCFtools v1.18⁸⁵. This resulted in a high-quality variation dataset comprising 10,622,425 SNPs, referred to hereafter as the “total variation dataset” (Set2).

To meet subsequent various analytical needs, we derived three additional SNP datasets from the total variation dataset: (1) LD-pruned dataset (Set3): This dataset removed SNPs in linkage disequilibrium (LD), with r² > 0.2 within 50-SNP windows, using PLINK v1.90b4.6⁸⁰. (2) Depth-consistent dataset (Set4): To mitigate batch and sequencing depth differences, this dataset was derived from the LD-pruned dataset, further excluding sites with FMT/DP < 10 and INFO/DP > 59,112 (twice the mean read depth). (3) Symmetric dataset (Set5): As the public available whole genome resequencing samples are mostly one individual per sampling location, to avoid the bias caused by unequal sample size, we initially selected only one individual with the highest sequencing quality from each Chinese population for genetic relationship analyses of global T. urticae. This dataset retained only the sample with the highest sequencing depth from each Chinese population (Supplementary Data 1), and included sites without missing (F_MISSING = 0). It should be noted that each instance involving individual variations includes additional filtering of monomorphic sites using BCFtools (AC == 0 || AC == AN).

Population structure and phylogenetics

The depth-consistent dataset (Set4) and symmetric dataset (Set5) were employed in population genetic analyses to study the population evolution of the invasive Chinese populations and global populations, respectively. The first ten principal components (PCs) of the genotypes were calculated using PLINK v1.90b4.6⁸⁰ with default parameters. The first two PCs were then visualized using ggplot2⁸⁶ to visualize genetic structure across individuals.

According to the PCA clustering, we categorized the Chinese populations into three groups. We calculated the F_ST and Dxy values between these groups in non-overlapping 500 kb windows across the entire genome using Pixy v1.2.7⁸⁷, and then averaged the results over all windows. To assess the effect of isolation by distance, we computed the Pearson’s correlation coefficient between geographic distance and population differentiation (F_ST/(1-F_ST)) for each pairwise population and determined the significance using the function ‘cor.test’ in the R package stats v4.3.1. Pairwise F_ST values among Chinese populations were also calculated with Pixy v1.2.7⁸⁷. Subsequently, we tested the significance of this linear relationship using Pearson’s correlation coefficient. We also compared the nucleotide diversity between the three Chinese groups with Wilcoxon rank test using the function ‘stat_compare_means’ in the package ggpubr v0.6.0.

We constructed a Neighbor-Joining (NJ) tree to infer phylogenetic relationships among the global T. urticae populations. We used one T. truncatus (SDjCo1) and one T. pueraricola (SCJ56-Tp) as the outgroup. The sequencing reads of these outgroup species were mapped to the T. urticae genome and performed SNP calling following the aforementioned method. The input DNA sequences were converted from the VCF file (Set5) using the script vcf2phylip v2.8⁸⁸. The NJ tree was built with 1000 bootstrap replicates using PHYLIP v3.697 (https://phylipweb.github.io/phylip/). One thousand resampled data sets bootstrapped over DNA sequences were created using SEQBOOT subroutine in PHYLIP phylip v3.697. Then the DNADIST subroutine was used to calculate the correlated pairwise genetic distances using the F84 distance model. The distances matrices (n = 1000) were used to construct NJ trees using the NJ tree NEIGHBOUR subroutine. Finally, a consense tree was generated using CONSENSE subroutine and visualized by iTOL (https://itol.embl.de/).

To infer the ancestral proportions for each individual, we used ADMIXTURE 1.3.0⁸⁹ with 10 replicate Markov chain Monte Carlo simulations for each K ranging from 1 to 20. The most likely number of genetic clusters was determined by examining the cross-validation error, with the smallest value representing the best estimate of K. The summation and graphical representation of genetic proportions were generated based on the average Q-matrices for all runs using CLUMPAK⁹⁰. This allowed us to visualize the genetic ancestry of each individual and assess the degree of admixture across different populations. We only performed these analyses for the filed-collected Chinese individuals, which were extracted from Set4.

Mitochondrial haplotype network

Haplotypes of the mitochondrial genome were constructed using the mitochondrial SNPs. Briefly, mitochondrial SNPs were extracted from the total variation dataset (Set2) and concatenated into a DNA sequence using vcf2phylip v2.8⁸⁸. Then, these sequences were used to construct haplotype network by PopART v1.7⁹¹ using the Median Joining Network method with default parameters (epsilon = 0)⁹².

Comparisons of SNP number and observed heterozygosity between Chinese and European samples

The SNPs in Set1 were further filtered using BCFtools to select loci that reached a depth of more than 10X in 60% of Chinese samples. These SNPs were utilized to compute the number of polymorphic SNP density within each 500 kb windows without overlap across genome for both European (n = 19) and Chinese samples (n = 25) separately using VCFtools v0.1.17⁸⁵. We calculated minor allele frequency (MAF) across all samples using PLINK v1.90b4.6⁸⁰. We divided MAF values into 5 bins, which are 0–0.1, 0.1–0.2, 0.2–0.3, 0.3–0.4 and 0.4–0.5 respectively. We then counted the number of SNPs in each bin for the Chinese and European samples, respectively, and visualized the results using ggplot2⁸⁶. The observed heterozygosity across all SNPs for each sample was calculated by PLINK v1.90b4.6⁸⁰. We compared the difference between the European and Chinese samples with Wilcoxon rank test using the function ‘stat_compare_means’ in the package ggpubr v0.6.0.

Gene flow inference

All the individuals of T. truncatus and T. urticae from China were extracted from the LD-pruned dataset Set3. The resulting 17,921 SNPs were analyzed using TreeMix v1.13⁹³ with a global set of rearrangements. Seven individuals of T. truncatus (Supplementary Data 6) were employed as the outgroup. Their SNP calling followed the aforementioned method with the T urticae genome used as reference. The migration events (m) were set to range from 1 to 10, and 10 replicated runs were conducted for each event under window sizes of 100 and 500 SNPs, respectively. The optimal number of migration events was determined using the Evanno method in OptM⁹⁴, with m = 6 identified as optimal. Subsequently, the introgression between populations was reanalyzed with 30 independent runs using this value and a consensus tree from 500 bootstraps, applying a window size of 500 SNPs in TreeMix. Finally, the tree with the highest likelihood was visualized, incorporating migration weights.

Identification of target-site resistance mutations

We scanned 30 target-site mutations across 10 genes associated with resistance to categorized to 12 different modes of action (Supplementary Data 7) to survey the frequencies of these resistance mutations in field populations in China and in the European samples. The genes included succinate dehydrogenase B and C (SdhB, SdhC), chitin synthase 1 (CHS1), the ATP synthase subunit c (ATPs), PSST homologue of complex I (PSST), glutamate-gated chloride channel subunit 1 (GluCl1), glutamate-gated chloride channel subunit 3 (GluCl3), acetylcholinesterase (AChE), voltage-gated sodium channel (VGSC), and mitochondrial cytochrome b (Cytb). Biallelic SNPs located in each of the pesticide-resistance-related genes were extracted from the total variation file Set2 and then annotated using SnpEff 4.3t⁹⁵ to determine the presence or absence of the missense mutations leading to resistant amino acids. The genotype frequencies of the resistance sites were scanned in 25 field populations in China. As G314D in GluCl1 and G326E in GluCl3 always appear together, we calculated D’ between the two sites using PLINK v1.90b4.6 to estimate the magnitude of linkage disequilibrium.

To determine whether resistance mutations arose from a single origin or from multiple origins, we constructed a haplotype network for each target gene. SNPs from exons and introns of each of the 10 target genes were extracted from the VCF file (Set2). The sequences of each target gene were subsequently phased into two haplotypes using the program Beagle v5.1⁹⁶ with the parameters of 20 burn-in iterations and 100 phasing iterations and concatenated using vcf2phylip v2.8⁸⁸. Then, haplotype networks for each target gene were built with PopART v1.7⁹¹, using the Median Joining Network method⁹² with default parameters (epsilon = 0).

To validate the haplotype network results, we also conducted the principal components analyses (PCA) for each gene using the SNPs for the above network analyses, but excluding the resistance mutation sites. In the case of a single original mutation, we would expect the resistant homozygotes would be clustered together in the PCA plots using SNPs within the resistance gene regardless of the field population they came from. The resistance sites were excluded to avoid confusion between identity by state and identity by descent¹⁶. In the case of independent evolution after invasion, SNPs near the resistance sites would be structured by population geography, not by resistance profile.

Genome-wide selection signals scans

To test for positive selection in Chinese populations, we performed selective sweep analyses in SweeD v4.0.0⁹⁷ with genome-wide SNPs (Set2). Composite likelihood ratios (CLR) were calculated in an average window of 1 kb across the genome by setting grid numbers according to chromosome lengths (number of grids = chromosome length/1000). We also calculated nucleotide diversity (π) and Tajima’s D to provide additional evidence of selection signals. Both indices were computed across the entire genome using a sliding window approach, with a window size of 50 kb and a step size of 5 kb. These calculations were performed using VCFtools v0.1.17⁸⁵ and VCF-kit v0.2.6⁹⁸.

Statistics and reproducibility

All statistical analyses were performed in R as described above. Comparisons between two groups with at least three independent replicates were conducted using the Wilcoxon rank test. The Mantel test was performed to assess isolation by distance after calculating Pearson’s correlation coefficient. The exact number of replicates (n) for each analysis is provided in the corresponding figure legends.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

J.T.S., L.C., X.Y.H., T.V.L., W.X., N.W., and X.F.X. designed the experiments. L.X.G., H.M.Z., X.N.S., Y.T.L. and J.J. performed the experiments. L.C., H.M.Z., and J.T.S. analysed the data. J.T.S., L.C. wrote the manuscript with input from N.W. and T.V.L.; All authors read and approved the final manuscript.

Peer review

Peer review information

Communications Biology thanks the anonymous reviewers for their contribution to the peer review of this work. Primary Handling Editors: Hannes Schuler and Michele Repetto. A peer review file is available.

Data availability

Whole-genome DNA sequencing data are deposited in the NCBI Short Read Archive (SRA) under BioProject PRJNA1162289. The source data for the main and Supplementary Figs. are available as Supplementary Data 19. All other data are available from the corresponding author on reasonable request.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s42003-025-08658-9.