Benzimidazoles.

Almost all samples were phenotypically resistant to benzimidazoles to some degree, reflective of their widespread use over many years and the ease by which T. circumcincta can develop resistance to treatment. Only the S_inbred isolate was fully susceptible, whereas MTci1 and MTci2 have each been previously suspected to harbour low levels of resistance [29], and the remaining isolates were considered to be resistant. Consistent among all comparisons that included a phenotypically well-defined resistant isolate with a phenotypically sensitive isolate was a peak on chromosome 1 that contained beta-tubulin isotype 1 (TCIR_10026770; gene position: 62.287–62.292 Mb), a gene well-established in its role in benzimidazole resistance. Three non-synonymous variants at positions P167, P198, and P200 of beta-tubulin isotype 1 are broadly associated with resistance in several nematode species [30–32]; here, we identify the F200Y allele at high frequency in all strongly resistant isolates, as well as the E198L variant in the Choi (RS³), Farm and MTci5 data, albeit at a lower frequency (Fig 3). Consistent with Choi et al., we did not find any haplotypes containing resistant alleles for both E198L and F200Y in any pooled sequencing dataset (S8 Table), further supporting the hypothesis that heterozygotes for both alleles are in trans- rather than cis- and that individuals homozygous for both resistance alleles are unviable [27].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Characteristics of genetic variation associated with resistance to benzimidazole treatment.

Resistance to benzimidazole treatment is strongly associated with variation in the beta-tubulin isotype 1 gene (TCIR_10026770). In most samples, high to moderate variant allele frequencies from whole genome sequencing data were observed for resistance variants F200Y and E198L, respectively. The S_inbred was the only wholly susceptible sample, whereas “susceptible” MTci1 and MTci2 have low levels of resistant variant F200Y. Resistance to higher levels of benzimidazole treatment has previously been associated with variation in the beta-tubulin isotype 2 gene (TCIR_10059730); here, we find the E198A variant only in the Farm 2 isolate.

https://doi.org/10.1371/journal.ppat.1012820.g003

In the predominantly susceptible MTci1 and MTci2 isolates, we see low levels of the F200Y variant, supporting the previous observations of suspected phenotypic resistance in these isolates and in agreement with the finding of low levels of the F200Y allele in MTci2 by Skuce and colleagues [29]. Additional beta-tubulin genes have been associated with resistance and can drive higher resistance levels than the isotype 1 variants alone [16,33]. Five beta-tubulin genes were identified in total in the tci2_wsi3.0 assembly, although not all have been associated with resistance previously. Here, we find the E198A variant in the beta-tubulin isotype 2 gene (TCIR_10059730) in the Farm 2 isolate only (Fig 3; freq_E198A = 12.7% and 15.4% in the pre and post-ivermectin-treatment samples, respectively). The presence of E198A in Farm 2 but not Farm 1 may explain the difference in benzimidazole treatment efficacy between the two farms (FECR_{benzimidazole} = 87.7% and 4.5% for Farm 1 and Farm 2, respectively), despite both having a very similar frequency of beta-tubulin isotype 1 resistant variants at both positions E198L and F200Y (Fig 3).

Levamisole.

Three isolates were phenotypically resistant to levamisole–RS³, MTci5 and MTci7 –however, only RS³ was directly exposed to levamisole immediately before sample collection. Despite the phenotypic history of the isolates, no shared peaks of genetic differentiation among the resistant isolates were identified that could be attributed to levamisole resistance. A single peak on the X chromosome unique to the S_inbred:RS³ comparison contained the acetylcholine receptor acr-8 (TCIR_10183160) (peak range: 1.5–5 Mb; gene position: 1.830–1.857 Mb), previously identified as associated with levamisole resistance (Fig 1B). In H. contortus, the S168T variant of acr-8 has been strongly implicated as being causal, having been functionally validated [34] and found in several levamisole-resistant populations around the world [16,35,36]; here, we identify the homologous variant S167T found at high frequency in RS³ (tci2_wsi3.0_chr_X, pos: 1,846,323, freq_S167T = 96.4%). This variant was, however, not found in the remaining levamisole-resistant isolates; the only other sample to contain it in a single sequencing read was Farm 2 (freq_S167T = 2%). The absence of the acr-8-associated peak in MTci5 and MTci7 suggests that other genetic variants may drive resistance; an alternative explanation, based on extensive experience with the cryopreservation of isolates, is that sample storage in liquid nitrogen can reduce the levels of levamisole resistance of an isolate post-recovery. We note that the levamisole resistance phenotype of the individual samples used was unknown. A key challenge to identifying these regions is that both isolates are also resistant to ivermectin and benzimidazoles, confounding the identification of drug-specific peaks. A review of previously described candidate genes associated with levamisole resistance highlighted this challenge (S7 Table); for example, the acetylcholine receptor subunit unc-29 (TCIR_10027440) was found to be closely located alongside beta-tubulin isotype 1 on chromosome 1 and, therefore, in these studies, if a genetic signal of differentiation associated with unc-29 were present, it would not be distinguished from the established signal associated with benzimidazole resistance. To do so, recombination mapping to identify independently segregating variants using sequencing of single-worms, phenotyped for levamisole sensitivity, could help map physically close variants associated with different drug classes. However, other subunits such as unc-63 (TCIR_10007740) on chromosome 1 at ~15.47 Mb and the levamisole-receptor genes lev-1.1 (TCIR_10117820) and lev-1.2 (TCIR_10117830) located on chromosome 4 at ~31.74 and ~31.81 Mb showed no clear genetic association with resistance.

Moxidectin.

Only the MTci7 isolate was moxidectin-resistant, so it is difficult to make robust conclusions from the analysis of a single isolate. Moxidectin is a macrocyclic lactone and belongs to the same drug class as ivermectin. Phenotypically, resistance to moxidectin typically confers resistance to ivermectin; however, resistance to moxidectin may occur independently of ivermectin resistance in C. elegans [37], a phenomenon also found in a recent field study of gastrointestinal nematodes on sheep farms [38]. More commonly, however, moxidectin can still be used to control nematode parasites in ruminants following a diagnosis of ivermectin resistance for a period of time [39], which may be due either to increased potency (a higher lipophilicity) or a different mechanism/inheritance of resistance [40]. A single peak on chromosome 3 (peak range: 15–20 Mb) found only in the MTci7:MTci2 comparison and not in the other ivermectin-resistant isolate comparisons is suggestive of a moxidectin-specific response. This peak contains the ortholog of hlh-4 (TCIR_00074870), the mutation of which has resulted in ivermectin and moxidectin resistance in C. elegans [37]. In C. elegans, hlh-4 is a transcription factor located within ADL neuronal cells that determines amphid neuron cell fate and influences the expression of a range of genes associated with ADL cells, including neurotransmitters, neuropeptides, ion channels, and electrical synapse proteins [41].

Moxidectin and ivermectin may share molecular targets, however, the genetic response to selection by each drug on those targets may differ. This could, in part, explain some differences in MTci2 comparisons with MTci7 relative to ivermectin-resistant isolates, such as the absence of the ivermectin peak on chromosome 5 or the more prominent peaks on chromosomes 4 (peak range: 85–88 Mb) and 5 (peak range: 73–76 Mb) in MTci2:MTci7 relative to MTci2:MTci5.

Ivermectin.

All comparisons involving ivermectin-resistant isolates, except for MTci7, revealed a common peak on chromosome 5 (consensus peak range: ~35–40 Mb). This peak was the only peak identified in the pre-to-post ivermectin treatment comparisons of the two farm populations, suggesting it was specifically an ivermectin-treatment-related response. To explore this region further, we focused on the Farm 1 population, in which we compared Fst from pooled sequencing (Fig 4A) with an independent experiment using ddRAD-seq to assess linkage disequilibrium (LD) among genotypes from individually re-sequenced larvae from the same pre- and post-treatment sample populations (Fig 4B). These analyses complemented each other, showing a region of differentiation and increased linkage in the post-treatment population at approximately 35–38 Mb, consistent with a reduction in diversity associated with drug-mediated selection. Two other peaks of LD were evident on chromosome 5, one of which corresponded with an Fst peak at approximately 73–76 Mb in the genome-wide Farm 1 vs Farm 2 post-treatment comparison and the MTci2 comparisons with MTci5 or MTci7 (Fig 2A and 2C). To further explore this peak of differentiation at 35–38 Mb, we assessed coverage (Figs 4C and Fig F in S1 Appendix), as well as gene, BUSCO, and repeat density in this region (Fig G in S1 Appendix); while there was increased variance in coverage in discrete regions of the locus, it did not explain the entire signal of differentiation, and there was no apparent bias in gene or repeat density suggesting it was an artefact of the assembly. Interestingly, variation in coverage was noted only between ~35–38 Mb between pre- and post-ivermectin samples on Farm 1, while the variation extended to 40 Mb on Farm 2, and, when post-ivermectin samples were compared (F2A:F1), the variation in coverage was primarily between 38–40 Mb, corresponding to the peak of genetic differentiation between the farms. These results collectively suggest that the locus under ivermectin selection lies between 35–38 Mb, although due to the lower coverage of RADloci between 34 and 35 Mb, this region cannot be excluded. To search for SNP alleles under selection by ivermectin on both farms, a Cochran-Mantel-Haenszel test was performed. This accentuated the significance of the 36–38 Mb region in particular (Fig H in S1 Appendix). Two SNPs were identified (Benjamini-Hochberg FDR 5%) as being significant following correction for multiple testing; Chr 5:38,037,930, peripheral to the ivermectin locus as defined here, and Chr 5:36,505,260, predicted to be an intron variant of TCIR_00147810, an orthologue of a H. contortus G protein receptor domain containing protein. However, given the availability of only two pairwise comparisons for the CMH test, additional experiments and replicates are needed to confirm whether these particular SNPs are important in ivermectin resistance, and whether multiple variants may individually (as for benzimidazoles or monepantel), or collectively confer ivermectin resistance. In our more conservative estimate of the ivermectin peak, ranging from 34–38 Mb, a total of 130 genes were identified in the region (S8 Table); considering that no strongly supported candidate genes were found, these results collectively lead us to conclude that this region has a novel driver of resistance.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Characteristics of genetic variation associated with ivermectin resistance.

A, B, C. Analysis of the major peak of differentiation on chromosome 5, comparing Fst between Farm 1 pre- and post-treatment pooled sequencing samples in 100 kb windows (A), linkage disequilibrium from ddRAD-seq of individual larvae from Farm 1 pre- and post-treatment populations (B), and relative sequencing coverage ratio between Farm 1 pre- and post-treatment pooled sequencing samples in 100 kb windows. D.E.F. Comparison of shared and unique signals genetic differentiation associated with ivermectin response on chromosome 5 of Teladorsagia circumcincta (D) and Haemonchus contortus (F). In both the D and F panels, data is coloured by the maximum Fst value in each experiment (low = white, high = blue). The middle panel (E) shows the relative position of 1-to-1 orthologous genes shared between the two species (n = 1,253 genes on chromosome 5, indicated by grey lines). The orthologous genes in the major QTL associated with ivermectin response in H. contortus are highlighted in red. G. Estimation of pgp-9 (TCIR_00132080) copy number expansion in resistant relative to susceptible isolates. The vertical dashed line is positioned at a copy number of one, representing the expected copy number ratio between ivermectin-susceptible isolates.

https://doi.org/10.1371/journal.ppat.1012820.g004

The shared peak between ivermectin-resistant isolates on chromosome 5 prompted us to compare it with the orthologous chromosome 5 of H. contortus. In previous works on H. contortus, we found a single peak associated with ivermectin resistance located at 37.5 Mb on chromosome 5, from which we proposed that overexpression of the bHLH transcription factor cky-1 was associated with resistance [16,42]. Comparison of Fst data highlighting the regions of differentiation between susceptible and resistant isolates of both T. circumcincta (Fig 4D) and H. contortus (Fig 4F), plotted together with 1-to-1 orthologs between the chromosomes of the two species (Fig 4E; grey lines), showed that the two major peaks in each species did not share genes and, therefore, are interpreted to be unrelated. However, a comparison of the orthologous genes within the H. contortus extreme quantitative trait locus (XQTL) peak with the T. circumcincta genome reveals two regions of interest (Fig 4E; red lines), with minor but significant peaks of differentiation between susceptible and resistant T. circumcincta. One of these peaks–found only in the Choi data–contains the T. circumcincta ortholog of cky-1 (TCIR_10182250). Considering that our previous hypothesis of cky-1 overexpression led to resistance in H. contortus, we had previously tested and confirmed that cky-1 was significantly higher expressed in the ivermectin-resistant MTci5 relative to the susceptible MTci2 isolate (see Fig 5E of [16]). We do not, however, see evidence of a peak of genetic differentiation between MTci5 and MTci2 around cky-1; this observation does not invalidate the previous hypothesis (i.e., that cky-1 is a crucial mediator of ivermectin resistance) but would require that cky-1 expression is influenced by selection in another region of the genome for it also to be involved in ivermectin resistance in MTci5. Alternative hypotheses include (i) that the two peaks in T. circumcincta associated with the single H. contortus peak represent a misassembly in the T. circumcincta genome and, if corrected, would resolve into a single peak in each, or (ii) that cky-1 at the end of the chromosome belongs in the prominent peak shared by all ivermectin resistant T. circumcincta isolates; however, careful examination of the Hi-C data, which should be sensitive to detecting misassemblies at this scale, does not support these alternate hypotheses.

The peak on chromosome 4 at ~85–88 Mb–prominent in the MTci2:MTci7 comparison and between-farm post-treatment populations–contains the ortholog of the p-glycoprotein gene pgp-9 (TCIR_00132080, 4:87.265–87.291 Mb). P-glycoproteins are members of the ATP-binding cassette (ABC) transporters, which function at the cellular membrane to pump foreign molecules out of the cell and are broadly implicated in resistance to chemotherapeutic drugs, including anthelmintics [43,44]. As such, the family of P-glycoproteins have been proposed to play a role in ivermectin resistance in several nematode species [45–47]. In T. circumcincta, constitutive over-expression of pgp-9 in triple-resistant MTci5 [48] and increased genomic copies of pgp-9 in the RS³ isolate relative to the S_inbred draft genome assembly and by qPCR of triple-selected adult males [27] have been described, both of which are consistent with positive selection to increase the rate by which anthelmintics, and possibly ivermectin in particular, are removed from cells. To assess pgp-9 copy number variation, we calculated the relative gene coverage genome-wide between susceptible and resistant isolates (Fig I in S1 Appendix), revealing pgp-9 as an outlier in multi-drug-resistant isolates. Comparatively, pgp-9 copies were highest in the resistant isolates in the following comparisons: MTci2:MTci7, followed by MTci2:MTci5 and RS³:Sinbred (Fig 4G), consistent with previous studies. The relative height of the pgp-9 peak to other peaks associated with ivermectin response is suggestive that pgp-9 variation is not the primary driver of ivermectin resistance in MTci5 and RS³ isolates but could be a strong candidate in MTci7. However, this cannot be attributed explicitly to ivermectin resistance in this study.

Additional smaller peaks of differentiation were identified between MTci2:MTci5, some of which contained genes previously associated with ivermectin resistance in C. elegans. For example, a gene associated with amphid channel morphology mutants resulting in ivermectin and moxidectin resistance in C. elegans, daf-6, is the ortholog of TCIR_00160170, found within the peak ~75 Mb along chromosome 5. It is unknown if variation in this gene, or others in minor peaks, are associated with resistant phenotypes in T. circumcincta, but warrant further investigation.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Summary of the main genomic regions and candidate genes associated with each anthelmintic class.

https://doi.org/10.1371/journal.ppat.1012820.t002

PacBio long-read sequencing.

Two PacBio long-read datasets were generated. The first long-read data were derived from whole genome amplified (WGA) DNA (REPLI-g Single Cell Kit, Qiagen) from two single adult male worms; the rationale of using WGA was to obtain micrograms of DNA for library preparation from the low amount of starting material (~10 ng per worm) while maintaining low genetic diversity from outbred samples. SMRTbell sequencing libraries prepared following the ‘20 kb Template Preparation Using BluePippin Size-selection System’ protocol were sequenced using 61 SMRT cells on the PacBio RSII platform, generating 6.5 million reads (N50 = 6.8 kb) representing ~63X coverage of the estimated 700 Mb genome (S1 Table). Initial assemblies using these data were highly fragmented, which we attributed to a high frequency of chimeric sequences in the WGA data. To address this, we made sequencing libraries from DNA obtained from a pool of 250,000 larvae, which were sequenced using 15 SMRT cells on the PacBio Sequel platform. In total, 7.4 million reads (N50 = 11 kb) were generated, representing ~74X raw coverage. Although these data contained significantly higher genetic diversity due to the large number of pooled larvae used, the issues with chimeric reads were resolved as WGA was not used.

Illumina Hi-C library preparation and sequencing.

Hi-C libraries were prepared using the Arima-Hi-C Kit for Animal Tissues (Arima Genomics) following the manufacturer’s instructions with 20 frozen adult female worms as input. An Illumina sequencing library was made using the Accel-NGS 2S Plus DNA Library Kit (Swift Biosciences) with a modified Arima Genomics protocol for low-input material. The library was sequenced on a single lane of Illumina HiSeq X10 using 150 bp paired-end chemistry, generating >150x coverage of the genome (S1 Table).

Genome assembly.

The PacBio Sequel data were initially assembled using Canu v.1.9 [62] (parameters: corMhapSensitivity = high corMinCoverage = 0). The high genetic diversity within the pooled larvae resulted in a highly haplotypic (BUSCO: 66% duplicated) and expanded assembly size (1.57 Gb), approximately twice as large as previous estimates (700.8 Mb; [27]). To reduce redundant haplotypes, we used purge_dups v.1.2.5 [63] to deduplicate the assembly, followed by Redundans v.0.14a [64] to re-scaffold using long reads after haplotypic sequences had been removed. Redundans and purge_dups were assessed separately and together on their ability to improve contiguity and decrease assembly size by eliminating redundant sequences while maximising complete and minimising duplicated BUSCOs.

Further assembly scaffolding was performed using the Illumina Hi-C data and Salsa2 [65]. Two rounds of scaffolding with Salsa2 were performed, with manual assembly curation between and after scaffolding. Manual curation was performed using pretext (https://github.com/wtsi-hpag/PretextMap; https://github.com/wtsi-hpag/PretextView) with HiGlass [66] contact maps as a guide to identifying misassembled regions and placing unassigned contigs and scaffolds into their chromosomal context. Where gaps in the assembly were created during curation, padding of 10 Ns were added. The Arima mapping pipeline (https://github.com/ArimaGenomics/mapping_pipeline) was used to generate data for curation, except that minimap2 rather than bwa-mem was used for mapping the Hi-C reads. Additional supporting analyses, including the calculation of PacBio read coverage (minimap2 -x map-pb), positions of repeats (Red [67]), assembly gaps (seqtk; https://github.com/lh3/seqtk), and telomeres (searching for a dimer of GCCTAA), were used to guide curation. The reference-aware scaffolder, RagTag v.2.1.0 [68], was used to assign the near-chromosome scaffolds into five autosomes and an X chromosome based on the chromosomal assembly of the closely related parasite H. contortus [69] after the first round of manual curation.

The assembly was further curated based on gene and sequence content. Before using Hi-C data, we observed that our original draft assembly contained single-copy BUSCO sequences that were missing from the Hi-C scaffolded assembly. Using an iterative and comparative approach, we recovered sequences from the draft assembly containing these missing BUSCO genes, merged them with the chromosomal assembly, and then removed redundant, contained contig and scaffold sequences (i.e., sequences that map within the boundaries of another sequence) by comparison using nucmer v4.0.0 [70]. This approach was followed with Blobtools [71] to remove potential contaminants, retaining sequences classified as ‘Nematoda’ by blast and unclassified sequences. Following this, the Arima mapping pipeline was repeated, and a final round of manual curation using Pretext and HiGlass was performed.

Finally, gaps were identified and filled using TGS-GapCloser v.1.2.1 [72], followed by polishing the assembly with long PacBio reads using arrow v.2.3.3 and three rounds of polishing using POLCA [73], the first using the Hi-C whole-genome sequencing reads and the second using RNA-seq reads (see below), followed by a final round using the Hi-C genomic reads. The rationale for using RNA-seq reads was to correct potential indels in the coding sequences. However, although successful, as a side effect, this approach caused errors in splice donor and acceptor sequences, leading to poor annotations. These were corrected (as was the annotation) with the final round of polishing with genomic reads.

Parasite material for genome annotation.

To support the genome annotation, we prepared RNA from L3, exsheathed L3, L4 (male and female), and adult (male and female) T. circumcincta MTci2. Briefly, L3 were obtained from coproculture from worm-free donor sheep faecal material containing eggs after incubation at 20°C for 10–14 days. Exsheathment was performed as previously described [74]. L4 and adult stages were harvested seven and 28 days post-infection, respectively, as previously described [75]. All samples were washed three times in PBS and subsequently snap-frozen in liquid nitrogen before storage at -70°C until RNA extraction.

RNA extraction.

Total RNA was extracted using a Quick-RNA tissue/insect Microprep kit (Zymo Research). Worm samples were homogenised in 800 μl of RNA lysis buffer by bead beating in ZR BashingBead Lysis Tubes (0.1 & 2.0 mm) using a Precellys 24 tissue homogeniser (Bertin Instruments) (x3 30 sec pulses at 5000 rpm). Total RNA was purified according to the manufacturer’s recommendations, with an on-column DNAaseI digest, and eluted in 15 μl nuclease-free water. RNA quantity and integrity were analysed using a NanoDrop One instrument (Thermo Fisher Scientific) and Agilent Bioanalyser 2100 system using RNA 6000 Nano kit. All RNA samples had a ribosome integrity number (RIN) ≥ 7. Total RNA was stored at -70°C before further use.

RNA-seq library preparation and sequencing.

All library synthesis and sequencing were performed by BGI Tech Solutions (Hong Kong). In brief, dual-indexed, strand-specific RNA-seq libraries were constructed from the submitted total RNA samples. The barcoded individual libraries were pooled and sequenced on a DNBseq G400 platform (paired-end, 2x150 bp sequencing), generating an average of 35.4 million reads per sample.

Iso-seq library preparation and sequencing.

Total RNA was isolated from T. circumcincta xL3, L4, adult male and adult female, as described above. Library preparation and sequencing were conducted at the University of Liverpool, Centre for Genomic Research. Samples were processed using a Teloprime full-length cDNA amplification kit v2 (Lexogen GmbH), following the manufacturer’s protocol. This involved first-strand cDNA synthesis by denaturing samples (1000 ng) at 70°C with primer RTP and, after cooling to 37°C, mixing with reverse transcriptase and incubating at 46°C for 50 min. Each sample was cleaned using the supplied columns, then ligated to a cap-dependent linker at 25°C for 3 h and cleaned with the same columns. The samples were incubated with second strand synthesis mix for 90 s at 98°C, 60 s at 62°C, and 72°C for 5 min, with a hold at 10°C before being column-cleaned again. Half of each sample was amplified for 15 cycles (one cycle of 95.8°C 30 s, 50°C 45 s, 72°C 20 min, followed by 14 cycles of 95.8°C 30 s, 62°C 30 s, 72°C 20 min) and cleaned with a column as before. The amplified cDNA products were made into SMRTbell template libraries according to the Iso-Seq protocol by Pacific Biosciences. Sequencing was performed on the PacBio Sequel System, and 1 SMRT Cell was run for each sample with a movie run-time of 600 min for each SMRT Cell.

Annotation.

The genome assembly was annotated using BRAKER3 [76]. Mapped T. circumcincta RNA-seq reads, and H. contortus protein sequences were provided to BRAKER3 as evidence. RNA-seq reads were mapped to the genome using HiSat2 (with—dta flag; [77]) followed by name sorting using samtools. Predicted H. contortus protein sequences were obtained from WormBase ParaSite release 18 PRJEB506; these protein sequences were chosen due to the relatively close phylogenetic relationship between T. circumcincta and H. contortus and because the H. contortus annotation has been extensively manually curated [15].

The initial exploration of the output of BRAKER3 revealed a large proportion of high-quality annotations, but some genes were not annotated by this pipeline despite the apparent support of RNA-seq mapping. We found that the BRAKER3 intermediate files produced by AUGUSTUS had greater sensitivity to detect these missing annotations but with lower specificity. To recover these genes, we used bedtools subtract to collect the missing genes from the AUGUSTUS annotation, after which gffread was used to remove single exon transcripts. This approach resulted in a substantial increase in BUSCO scores after merging the BRAKER3 and recovered AUGUSTUS features.

Manual curation.

We used the genome annotation to further curate the assembly. Coding sequences annotated on linkage groups and unplaced scaffolds were aligned to the chromosomal scaffolds using exonerate (—model protein2genome—percent 50) [78]. Scaffolds and annotations that were syntenic with chromosomal scaffolds were subsequently removed from the assembly. This approach was further supported by removing scaffolds containing duplicated BUSCO sequences between scaffolds and chromosomes. Finally, manual curation of genes of interest (for example, ion channels and genes associated with resistance to at least one anthelmintic or those involved in vaccine development) and some surrounding regions was performed using Apollo 2 [79]. Using RNA-seq and Iso-seq mapping tracks as evidence, approximately 3,200 genes were edited in the first instance; however, this is an ongoing effort for which annotation updates will be made. These curated annotations were reintroduced into the BRAKER3+AUGUSTUS annotation described above after removing redundant and erroneous gene models using bedtools subtract and AGAT [80]. AGAT was further used to add start and stop annotations (agat_sp_add_start_and_stop.pl) and to perform analyses using the longest isoform per gene (agat_sp_keep_longest_isoform.pl).

Genomes compared.

We compared our assembly and annotation against two previously published T. circumcincta assemblies. The first draft genome assembly for T. circumcincta was published in 2017 (Assembly name: T_circumcincta.14.0.ec.cg.pg; assembly accession: GCA_002352805.1) [27], derived from a partially inbred anthelmintic susceptible parasite isolate from New Zealand, hereafter referred to as the WASHU genome. The second assembly was derived from the T_circumcincta.14.0.ec.cg.pg assembly but curated and scaffolded using Hi-C by the DNAzoo Consortium to achieve chromosome-scale scaffolds [27,81,82] (https://www.dnazoo.org/assemblies/Teladorsagia_circumcincta; assembly name: T_circumcincta.14.0.ec.cg.pg_purgehaplotigs_HiC.assembly). These three T. circumcincta assemblies were also assessed against the chromosomal assembly of the H. contortus ISE isolate [69] and the C. elegans N2 reference genomes.

Repeat Content.

Repetitive sequences were assessed in the three T. circumcincta genomes. For each genome, RepeatModeler v2.0.1 was used to detect and classify repeats using default parameters, followed by RepeatMasker 4.1.0 (parameters: -s -html -gff -small -poly) to annotate the genome and summarise the repeat classes. Repeat density per Mb was calculated using a custom script and plotted by chromosome (Fig G A in S1 Appendix).

Genome completeness.

A comparison of intermediate and complete assembly versions was performed using BUSCO v.5.6.1 [83] (parameter:—genome) using the nematoda_odb10 lineage reference datasets. Completed assemblies with annotations were further assessed by BUSCO using predicted translated proteins (parameter:—protein). Genome assemblies were also evaluated using compleasm [84]. Gene density per Mb was calculated using a custom script and plotted by chromosome (Fig G B in S1 Appendix). In addition, compleasm output was used to plot the number of busco genes per Mb, indicating whether they were identified as a single copy or duplicated in the tci2_wsi3.0 assembly (Fig G C in S1 Appendix).

Protein comparisons.

Protein sequence orthology between genome assemblies was assessed using Orthofinder v.2.3.3 [85]. Predicted protein sequences were obtained from the genome and annotation using gffread, from which the longest protein sequence per gene was chosen to assess using Orthofinder. To enhance the comparison between WASHU and tci2_wsi3.0 assemblies, we provide the orthogroup mapping between predicted protein sequences from the two assemblies in S11 Table. Functional annotations for the protein sequences were inferred using InterProScan v.5.57–90.0 against all databases (—goterms—iprlookup). The output of InterProScan was collated, and the functional annotations were added to the GFF using AGAT [80] agat_sp_manage_functional_annotation.pl. To support further analysis using these data, we provide GO terms associated with tci2_wsi3.0 transcripts in S12 Table.

Parasite material for population genetic analyses.

Two UK field populations of T. circumcincta were obtained from commercial sheep flocks to investigate genetic markers of resistance using the faecal egg count reduction test (FECRT) in 2015 and 2016. Here, resistance is defined as a reduction in arithmetic mean faecal egg count (FEC) of less than 95%, with the lower 95% CI less than 90% [86]. Farm 1, sampled in August 2016, was located in SW England and has been previously described [87,88]. This flock (‘Farm 8’ in Glover et al. [88]) had resistance to multiple anthelmintic classes diagnosed in 2013; FECRT_{benzimidazole} = 83.7%, FECRT_levamisole = 92.5%, and FECRT_ivermectin = 93.0% reduction in faecal egg output. Subsequently, a second flock was introduced to Farm 1 to form a single, larger flock. Following this, for the current study, a FECRT was performed in 2016. Briefly, sheep were individually weighed to the closest 0.5 kg before being treated with 0.2 mg/kg ivermectin (Noramectin 0.08% oral solution for sheep, Norbrook, UK), with the dose rounded up to the nearest 0.2 ml [87]. Eggs were counted using a modified Mini-Flotac method, with a detection limit of five eggs per gram (epg) [87]. For Farm 2, located in central Scotland, a FECRT was performed in July 2015. Sheep were treated with 5 mg fenbendazole/kg (10% Panacur, MSD Animal Health Ltd, UK), 0.2 mg ivermectin/kg (Oramec, Boehringer Ingelheim Animal Health, UK), and 7.5 mg levamisole hydrochloride/kg (3% Levacur, MSD Animal Health Ltd, UK), being individually weighed and dosed according to weight, and a FEC was performed using a cuvette method with a detection limit of 1 epg [89]. On both farms, faeces were collected pre- (on the day of treatment) and 14 days post-ivermectin treatment of sheep, and L3 were recovered following coproculture; for Farm 1, faeces were stored at 8°C for one week and then cultured at 20°C for 14 days, while for Farm 2 faeces were cultured at 20°C for ten days. For this study, individual L3 were identified at the species level using PCR of the ITS2 region, and the mean FECs were adjusted to calculate an approximate percentage faecal egg count reduction (FECR%) for T. circumcincta as previously described [90]. Arithmetic means and standard deviations were calculated using Microsoft Excel. The eggCounts R package v2.4 [91], fecrt_stan, with the unpaired model was used to calculate the FECR% and highest posterior density interval. This model was chosen as it allowed unequal group sizes and individual correction factors for each FEC. For Farm 1, the model accounted for zero inflation, arising from low counts that fall below the FEC detection limit and are falsely observed as zero. For Farm 2, the FEC method had a detection limit of 1 epg, hence, no egg counts were taken to be genuinely zero, and zero inflation adjustment was not used. Data for the FECRT are shown in S1 Table and Fig C in S1 Appendix.

A collection of archival samples biobanked at the Moredun Research Institute were also sequenced. These samples included two UK T. circumcincta isolates previously characterised for their susceptibility to drug treatment; MTci1, isolated in 1979 from Moredun Research Institute pastures (drug-sensitive, although suspected low-levels of benzimidazole resistance) and MTci2, the same isolate as used for the genome assembly. Two multi-drug-resistant isolates were also sequenced. MTci5 (benzimidazole, levamisole and ivermectin-resistant) was originally isolated in 2002 from a farm in central Scotland following a diagnosis of triple anthelmintic resistance in an experimental Suffolk flock a year previously [92,93], with subsequent diagnosis of a lack of persistency of moxidectin in a commercial ewe flock grazed on the same pastures [94]. MTci7 (benzimidazole, levamisole, ivermectin, and moxidectin-resistant) was isolated from a large upland sheep farm in Scotland in 2007 following diagnosis of moxidectin resistance by FECRT, subsequent to detection of a lack of persistency of moxidectin two years previously [95]. All samples consisted of hundreds of adult or L4 worms, sampled before or after drug treatment, as described in S2 Table.

Illumina library preparation and sequencing.

Genomic DNA was extracted from the UK field populations (Farm 1 and Farm 2). Initially, DNA lysates of individual larvae were prepared using DirectPCR Lysis Reagent (Cell; Viagen Biotech) supplemented with DTT (Thermofisher Scientific) and Proteinase K (Fungal; Invitrogen), and speciation PCRs performed using species-specific ITS2 single- and multi-plex PCRs to identify T. circumcincta L3 from other strongyle larvae, as previously described [96]. Ninety-one T. circumcincta L3 were pooled, using half of the total lysate volume, and genomic DNA was extracted using a phenol-chloroform method supplemented with glycogen during the final ethanol precipitation to enhance the recovery of the low DNA quantity present, as previously described [90]. The final precipitated genomic DNA was resuspended in 10 μl of EB buffer (Invitrogen). Illumina sequencing libraries were prepared, followed by 75 bp paired-end sequencing of each library on an individual lane of Illumina HiSeq 4000. For Farm 2, a technical replicate of genomic DNA extraction, library preparation and sequencing was performed, and these data are referred to as 2A and 2B (S2 Table and Fig E in S1 Appendix).

To assess linkage disequilibrium, we used double digest restriction-site associated DNA sequencing (ddRAD-seq) on sixty-two individual L3 (30 pre- and 32 post-ivermectin treated L3) from Farm 1. These L3 were a subset of those included within the larger number pooled for whole genome sequencing above. Briefly, half the volume of DNA from individually lysed larvae (DirectPCR Lysis Reagent (Cell; Viagen Biotech), DTT, Proteinase K) was whole genome amplified (REPLI-g Single Cell Kit, Qiagen) before digestion with two restriction endonucleases, MluCI and NlaIII (New England Biolabs), as previously described [90]. Barcoded adaptors [97] (S3 Table) were ligated to 250 ng of digested whole genome amplified DNA. Pooled sample DNA libraries were size selected (225–325 bp) using a Pippen Prep (Sage Science), followed by the addition of Illumina adapters by PCR using KAPA HiFi (KAPA Biosystems) PCR for eight cycles. Genomic libraries were sequenced with 125 bp paired-end reads on an Illumina HiSeq 2000v4.

Genomic DNA was extracted from the archival drug-resistant isolates using the Isolate II DNA extraction kit following the manufacturer’s instructions, with a final elution in 50 μl buffer. Illumina sequencing libraries were prepared, followed by 150 bp paired-end sequencing on a single lane of Illumina NovaSeq 6000 (S2 Table).

Mapping Illumina reads to reference genomes.

In addition to the sequencing data generated for this study and described above, we also reanalysed pooled sequencing data from a semi-inbred drug-susceptible (S_inbred) and a multidrug-resistant, backcrossed isolate (RS³) originally derived from a resistant field isolate; the founding isolates were collected from farms in New Zealand. These data were originally generated to map drug resistance variation as described by Choi and colleagues [27] and are available from ENA BioProject PRJNA72569 (RS³: SRR4026813; S_inbred: SRR4027077) (S2 Table).

Sequenced and ENA-derived Illumina reads were quality checked using FastQC v0.12.1 (https://www.bioinformatics.babraham.ac.uk/projects/fastqc/) and MultiQC v1.17 [98]. Sequencing reads were mapped to a reference genome using a Nextflow mapping pipeline (mapping-helminth/v1.0.8). Briefly, the pipeline first converted raw reads into an unmapped BAM (uBAM) file using FastqToSam and subsequently processed them using MarkIlluminaAdapters from GATK v.4.1.4.1 before using minimap2 v.2.16 [99] to map the sequencing reads to the T. circumcincta tci2_wsi3.0 reference genome. Mapped reads were sorted, and duplicate reads were marked using samtools and sambamba [100], respectively. Mapping statistics were generated using samtools flagstats and summarised using MultiQC.

Comparison of coverage between farm samples over chromosome 5.

To compare coverage between pre- and post-ivermectin samples along chromosome 5, the depth per site was calculated using samtools v1.17 depth (-q 30 -Q 30 -a), requiring both the base and mapping quality to be at least 30, and for all sites to be included in the output, even if zero coverage, from which the average depth was calculated in 100 kb windows along the chromosome. Note that split and secondary hits had already been removed using a combination of sambamba and awk. Next, the coverage was standardised by using the median chromosome 5 coverage. Finally, the ratio of the post-ivermectin:pre-ivermectin sample was calculated.

Within and between sample analyses of genetic differentiation.

To calculate between-sample genetic differentiation, we used Grenedalf v0.3.0 fst [101] to calculate an unbiased-Nei measure of Fst in 100 kb sliding windows. We ran calculations separately for the “farm”, “choi”, and “strain” data using mapped bam files as input, keeping the run parameters consistent (--method unbiased-nei--sam-min-map-qual 30--sam-min-base-qual 30--filter-sample-min-count 2--filter-sample-min-coverage 10--filter-sample-max-coverage 100--window-type sliding--window-sliding-width 100000--write-pi-tables) but to account for the different estimated pool sizes of each sample group (--pool-sizes = 800, 180, and 200 for the “choi”, “farm”, and “strain” analyses, respectively). Pairwise Fst data per 100 kb for all sample combinations within each of the “farm”, “choi”, and “strain” groups are presented in S13, S14, and S15 Tables, respectively.

To assess the impact of coding sequence variation and determine individual variant frequencies, variants were identified using bcftools (v1.14) mpileup (-B -q 30 -Q 30—ff DUP -d 100) followed by bcftools call (-mv -O v). Putative coding variation was assessed using SnpEff (v5.2a) [102].

To identify whether any SNP alleles were under selection by ivermectin on both UK farms, a Cochran-Mantel-Haenszel test was performed using PoPoolation 2 cmh-test.pl (v1201, “popoolation2-code” version which provided the log odds ratio) (—min-count 3—min-coverage 10—max-coverage 2%—population 1–2,3–5). The Farm 1 pre- and post-treatment pairwise comparison was compared with the Farm 2A pre- and post-treatment pairwise comparison. P values were filtered using both the FDR (Benjamini-Hochberg), choosing an FDR of 5%, 1% and 0.01%, and also by using the Bonferroni method. Finally, the proportion of SNPs ranked within the top 1% of the lowest p-values, which also had a probability of a reduction in the reference allele in the post-treatment samples relative to the pre-treatment samples (log odds ratio < 1), were calculated in 1 Mb windows along the chromosomes and plotted using ggplot2.

Gene copy number variation.

To determine the relative copy number difference of genes between susceptible and resistant isolates, we first determined the read coverage of all CDS exons within the annotation using bedtools (v2.29.0) multicov. In R, the median CDS coverage per gene was determined, from which a ratio of resistant-to-susceptible gene coverage was calculated. To determine outliers of coverage ratio, thereby identifying genes with significantly higher or lower coverage in the resistant relative to susceptible isolate, we ranked genes by the susceptible isolate coverage and plotted the coverage of the resistant isolate using this gene order. This approach was used to assess copy number variation in pgp-9 (TCIR_00132080).

ddRAD-seq analyses.

Sequencing data were de-multiplexed using Stacks v2.5 process_radtags (-r -q), followed by mapping to the tci2_wsi3.0 assembly using the mapping-helminth v1.0.8 pipeline described above. Variants were called using GATK HaplotypeCaller v4.1.4.1 (parameters: -ERC GVCF—min-base-quality-score 30—disable-read-filter NotDuplicateReadFilter) per sample, followed by GATK CombineGVCFs to compile all 62 sample sets into a single vcf file. The vcf was filtered using GATK VariantFiltration to keep only variants successfully genotyped in at least 30 individuals (—filter-expression “AN < 60”) and with a mapping quality greater than 30 (—filter-expression “MQ < 30”). A threshold of 30 individuals was chosen to allow for the inclusion of RAD sites, which were present in the pre-treatment or the post-treatment sample population but lost in the other. A prior analysis using an earlier version of the assembly had identified that two individuals–one in the pre-ivermectin sample and the other in the post-ivermectin sample–were identical, and thus, these individuals were removed from further analysis (see [90] for more details).

Linkage disequilibrium within each sample was analysed along chromosome 5 and between chromosomal scaffolds using PLINK v1.9 [103]. Inter-chromosomal r² was calculated using plink—r2 inter-chr with-freqs, allowing 25% missing individuals (—geno 0.25) and a minimum maf (—maf 0.05). Sites were filtered only to include one variant every 100 kb (—bp-space 100000), and all r² values were retained (—ld-window-r2 0). To calculate r² along chromosome 5 (—chr tci2_wsi3.0_chr_5), the same command was used (—r2 with-freqs), but increasing the number of sites to a maximum of one site every 1 kb (—bp-space 1000), and setting the maximum ‘distance’ between two SNPs at 1 Mb (—ld-window-kb 1000) or 5 Mb, and with no more than 999 other SNPs between them (—ld-window 1000). Finally, the output data were restricted to only r² values of 0.001 or higher. The decay of linkage disequilibrium over chromosome 5 was investigated. For this, a script written by Fabio Marroni was used [104]. The distance between variants (in bp) to the half-decay value (r² = 0.5) was calculated. Next, the distance to the mean sample inter-chromosomal r² value was calculated. As these were slightly different for the pre- and post-ivermectin samples, the distance to an r² value of 0.05 was also calculated.