We assembled the genome of A. lycopersici into seven scaffolds of cumulative length 32.53 Mb, of which 99.98% is represented on scaffolds 1–5 of lengths 12.44, 10.50, 3.66, 3.57 and 2.36 Mb, respectively. The remaining two scaffolds are each <6 kb in length, in addition to a mitochondrial genome scaffold. The observed assembly length is similar to the length estimated by a k-mer analysis with genomic sequence reads (34.81 Mb). Separate genome completeness estimates with CEGMA (Parra et al., 2007) and BUSCO (Simão et al., 2015) located 90.7% and 86.0% of the expected core eukaryotic genes, respectively; these values are within the same range as those for T. urticae, the only other sequenced chelicerate herbivore, and for which a high-quality Sanger assembly is available (95.16% and 92.07%, respectively). As an additional assessment of completeness, we generated a de novo assembly of the A. lycopersici transcriptome using deep, paired-end Illumina RNA-seq reads derived from mixed sex and developmental stages, and aligned it to the genome sequence. We found that 98.2% of transcript contigs could be located on the reference sequence. Of the remaining 243 unplaced transcript sequences, only eight had similarity to known arthropod sequences; the others had homology to bacterial, fungal, or plant sequences, or lacked homology to sequences in existing databases.

Annotation of the A. lycopersici genome by automated methods, coupled with extensive manual curation, revealed only 10,263 protein-coding genes. As assessed against other mite genomes, including T. urticae, Dermatophagoides pteronyssinus (the European house dust mite) (Waldron et al., 2017), and Metaseiulus occidentalis (a phytoseiid predatory mite) (Hoy et al., 2016), as well as the Drosophila melanogaster and human genomes, several features of genic organization in A. lycopersici stand out (Table 1). The fraction of the genome comprising coding sequence is highest in A. lycopersici, and the distance between genes is the lowest. Associated with the compact genic landscape of A. lycopersici (Figure 2 and Figure 2—figure supplements 1–6), the percentage of the genome consisting of transposable elements was merely 1.54%, which is more than fourfold less than that observed in several other mite genomes, or in the insect D. melanogaster (Figure 2—figure supplement 1, Supplementary file 1 — ‘Table S1’ Tab). Nevertheless, sequences homologous to the major classes of transposable elements, such as DNA transposons, including Helitrons, as well as both long terminal repeat (LTR) and non-LTR retrotransposons, were detected (Supplementary file 1 — ‘Table S1’ Tab and ‘Table S2’ Tab). Across the A. lycopersici genome, extended regions of low genic composition and high TE density were not observed (Figure 2—figure supplement 2), consistent with the purported holocentric chromosome architecture (lack of regional centromeres) of eriophyoid mites (Helle and Wysoki, 1996).

We also observed that the A. lycopersici genome has only 3057 introns in coding sequences (CDS introns), which is more than an order of magnitude fewer than the 44,881 in the 90 Mb T. urticae genome, and the 35,841 in the 70.8 Mb D. pteronyssinus genome. Strikingly, nearly 84% of A. lycopersici protein coding genes were intronless, which is more than threefold higher than observed for the other mite species we analyzed, and more than fivefold higher than for D. melanogaster (Table 1). To further investigate the dynamics of intron evolution, we evaluated patterns of intron gain and loss in orthologous genes among A. lycopersici and 17 other animal genomes using the Malin analysis pipeline (Csurös, 2008; Figure 2, and Figure 2—figure supplements 3 and 4, and Supplementary file 2). At 29,447 conserved intron sites (Figure 2a), A. lycopersici has a mere 207 introns. This is an ~11 fold reduction from that seen in the species with the next lowest counts, the European house dust mite D. pteronyssinus, at 2292. Apart from A. lycopersici, Acari intron loss rates were broadly similar to those observed for other arthropods, except for M. occidentalis, for which high rates of both intron loss and gain were apparent, a finding previously reported (Hoy et al., 2016). However, the rate of intron loss in A. lycopersici was higher than observed in M. occidentalis (Figure 2b), and in contrast to M. occidentalis, intron gains were minimal (Figure 2—figure supplement 4). The only evidence for retention of the minor spliceosome in A. lycopersici comes from the presence of a single canonical U12 (minor) intron in the gene aculy03g00270 that encodes an ultra-conserved calcium channel (splice sites AT-AC in intron one of length 12.5 kb). Splicing of this large intron is supported by RNA-seq read alignments, and the orthologous intron one of the T. urticae orthologue of this gene is one of the three U12 introns documented previously in T. urticae (Grbić et al., 2011).

Although relatively few conserved introns are present in the A. lycopersici genome, they exhibit a bias toward 5’ gene ends (Figure 2—figure supplement 5), and compared to most arthropods, the median intron length is larger (Table 1 and Figure 2—figure supplement 6). In a single copy (orthologous) gene set for which introns were lost in A. lycopersici, but conserved in five other closely related or high-quality mite or insect genomes (see Materials and methods), the impact of intron loss on A. lycopersici-encoded protein sequences was generally minimal. In fact, in the respective protein sequences spanning 97 of 100 A. lycopersici-specific intron loss events (97%), multi-species alignments did not reveal insertions or deletions (indels) of amino acid residues (e.g. Figure 2c, and Supplementary file 1 — ‘Table S3’ Tab and Supplementary file 3); for the remaining few cases (3%), the respective sites of loss events in A. lycopersici were coincident with the gain or loss of one or several amino acid residues (e.g. Figure 2d). Within this gene set, similar findings were apparent for the larger number of A. lycopersici intron losses as compared to intron sites conserved between the two closest relatives (D. pteronyssinus and T. urticae; Supplementary file 3). Despite striking examples of intronless genes arising from the loss of multiple conserved introns, as for aculy03g01320 (Figure 2c), some A. lycopersici genes have both lost and retained arthropod conserved introns (i.e. aculy02g00250, aculy03g02140, and aculy01g28080, Supplementary file 3).

As revealed by the clustering algorithm implemented in the CAFE software (Han et al., 2013), A. lycopersici exhibits one of the highest rates of gene family contractions (1725), and by far the lowest rate of gene family expansions (206), among the 18 metazoans we analyzed (Figure 3; input data for the analysis are provided in Supplementary file 4 and Supplementary file 5). It also has the lowest average expansion per gene family (Supplementary file 1 — ‘Table S4’ Tab). Of the 105 gene families that were identified as ‘rapidly evolving’ in A. lycopersici, only four – as represented by orthogroups (OGs) OG0000007 (containing an Asteroid domain: IPR026832), OG0000546 (containing a Major Facilitator Superfamily, or MFS, domain: IPR011701), OG0000583 (containing a Troponin domain: IPR001978), and OG0002260 (hypothetical proteins) – were identified as expanding. The remaining 101 families were all identified as contracting (Supplementary file 1 — ‘Table S5’ Tab). Six of these contracting families did contain more than 10 members in A. lycopersici (OG0000000, containing a Zinc finger C2H2-type domain: IPR013087; OG0000003, containing a Homeobox domain: IPR001356; OG0000005, containing a Serine protease, trypsin domain: IPR001254; OG0000014, containing a Cytochrome P450 domain: IPR001128; OG0000015, containing a G-protein-coupled receptor, rhodopsin-like domain: IPR000276; G0000025, containing a Homeobox domain: IPR001356) and, except for OG0000014 containing members of the P450 family, which is known to have only few orthologous relationships (Feyereisen, 2011), on average 72.2% of retained A. lycopersici genes had an orthologue in the majority of chelicerate species (Supplementary file 1 — ‘Table S6’ Tab). Further, among the 101 rapidly contracted gene families we identified families previously implicated in mite and insect xenobiotic detoxification (Dermauw et al., 2013a; Dermauw et al., 2013b; Snoeck et al., 2018; Van Leeuwen and Dermauw, 2016) – carboxyl/choline esterases (CCEs: OG0000021 and OG0001201), cytochrome P450 monooxygenases (CYPs: OG0000014, OG0000030 and OG0000052), glutathione-S-transferases (GSTs: OG0000102, OG0000124), short-chain dehydrogenases/reductases (SDRs: OG0000096), ATP-binding cassette (ABC) transporters (ABCs: OG0000051 and OG0000109) and MFS proteins (OG0000029, OG0000071, OG0000099, OG0000187) (Supplementary file 1 — ‘Table S5’ Tab and ‘Table S7’ Tab). Given the role of these families in herbivory and host plant use (Després et al., 2007; Heckel, 2014; Van Leeuwen and Dermauw, 2016), we analyzed a selection of these gene families in detail (see the following sections).

We also found 315 orthogroups with no members in A. lycopersici but at least one member in all other arthropod species. This is the highest number of absent orthogroups of all arthropods included in our analysis, is ~2-fold more than those lacking in D. pteronyssinus (171), and more than threefold those absent in T. urticae (101) (Supplementary file 1 — ‘Table S8’ Tab). A gene ontology (GO) enrichment analysis for D. melanogaster members within these conserved arthropod orthogroups without A. lycopersici members revealed that N-acetylglucosamine metabolic process (GO:0006044), transferase activity (GO:0016740) and Golgi apparatus (GO:000579) were the most highly significantly enriched GO terms within the Biological Process, Molecular Function and Cellular Component GO categories, respectively (Supplementary file 1 — ‘Table S9’ Tab). Lastly, we found that 427 D. melanogaster essential genes (Aromolaran et al., 2020) coded for members of 390 orthogroups. Forty-eight of these essential orthogroups did not have members within the Acariformes, the mite superorder comprising A. lycopersici, D. pteronyssinus, and T. urticae, while 21 (5.4%) orthogroups were absent in A. lycopersici but present in other acariform mites (Supplementary file 1 — ‘Table S10’ Tab).

Furthermore, in a number of cases, orthogroups absent in A. lycopersici harbor conserved genes with potential roles in the development of tissues or structures that are absent or modified in the russet mite relative to other chelicerates or insects (see also Discussion, and Results section, ‘Loss of highly conserved transcription factors’). For instance, orthologues of Drosophila unkempt, a known developmental regulator, and Drosophila dachs, essential for appendage growth, are both absent in A. lycopersici but present in all other arthropods (OG0002898 and OG0006002, respectively). Dachs is known to interact with four-jointed (Buckles et al., 2001), which is also absent in A. lycopersici, even though it is present in all insect and chelicerate species included in our analysis (OG0003305). Finally, fat belongs, together with dachs and four-jointed, to the Fat/Hippo pathway and plays a key-role in tissue proliferation and development in both invertebrates and vertebrates (Simon et al., 2010). Although dachsous, another player in this pathway, is present (aculy04g02000 in OG0001018), a fat orthologue could not be identified in A. lycopersici while this orthologue was found in other acariform mites (OG0000383, Supplementary file 1 — ‘Table S7’ Tab).

We curated the A. lycopersici genome for sequences encoding established detoxification enzymes (Després et al., 2007; Heckel, 2014; Van Leeuwen and Dermauw, 2016) including GSTs, CCEs, and CYPs. In A. lycopersici, detoxification gene families are especially reduced, with a mere 4 GSTs, 8 CCEs, and only 23 CYPs (Table 2, Figure 4a, and Figure 4—figure supplements 1, 2 and 3; Van Leeuwen and Dermauw, 2016). In particular, the number of GSTs and CCEs is remarkably low (see Discussion). This finding was corroborated by mining of the A. lycopersici transcriptome assembly (the 4 GSTs and 8 CCEs present in the genome assembly were also present in transcriptome assembly, with no other transcript contigs with homology to GSTs or CCEs identified). Of note, half of the GSTs and almost all (7 out of 8) CCE genes in A. lycopersici are evolutionarily conserved across chelicerates or arthropods (Figure 4—figure supplements 1 and 2). We also examined transporters of the ABC family and MFS proteins that have been implicated in detoxification responses in arthropod species, although transporters in both of these families have diverse other roles as well (de la Paz Celorio-Mancera et al., 2013; Dermauw et al., 2013a; Dermauw et al., 2013b; Dermauw and Van Leeuwen, 2014; Govind et al., 2010). In contrast to genes encoding ‘classic’ detoxification enzymes like CYPs, CCEs, or GSTs, dramatic reductions in ABC transporter genes were not observed. For example, A. lycopersici has 9 ABCC and 16 ABCG transporters, while 22 and 2 are present in M. occidentalis and 39 and 23 are present in T. urticae, respectively (Table 2, Figure 4—figure supplement 4). Further, in contrast to the trend for contractions of the classic detoxification gene families, we also observed two A. lycopersici expansions - comprising three orthogroups, OG0000024, OG0000546, and OG0006109 - of the MFS, which is involved in membrane-based transport of small molecules (Figure 4b, Figure 4—figure supplement 5; Pao et al., 1998; Yan, 2015).

To see if A. lycopersici’s specialized lifestyle has had a notable impact on chemoreception, we also exhaustively mined and annotated the A. lycopersici genome for gustatory receptors (GRs), degenerin/epithelial Na+ channels (ENaCs), ionotropic receptors (IR) and transient receptor potential (TRP) channels. Members of these four families have been previously documented to play important roles in sensing environmental (chemical) cues in other arthropod species (Damann et al., 2008; Hoy et al., 2016; Ngoc et al., 2016; Robertson et al., 2003; Rytz et al., 2013; Whiteman and Pierce, 2008). The GR family, which contains seven transmembrane spanning regions (Touhara and Vosshall, 2009) and is linked to the detection of sweet and bitter compounds (Silbering and Benton, 2010), was the most strongly reduced, with only two of these genes identified (Figure 4c, Figure 4—figure supplement 6), as opposed to the 447 intact GRs reported in T. urticae (Ngoc et al., 2016). Further, only four ENaCs are present in the A. lycopersici genome (Figure 4d, Figure 4—figure supplement 7). Members of this family have recently been shown or suggested to be chemoreceptors for diverse compounds in insects and mites, but some family members likely have highly conserved roles in acid sensing (Ben-Shahar, 2011; Silbering and Benton, 2010), as well as in the perception of mechanical or osmotic cues (Ben-Shahar, 2011; Zelle et al., 2013). Of the two ENaCs likely to play these conserved roles in T. urticae, one is in a well-supported clade with a single ENaC in the tomato russet mite (aculy04g09940) (Figure 4 , Figure 4—figure supplement 7).

The IR family, which has been linked to odorant detection (Joseph and Carlson, 2015), humidity and temperature sensing in D. melanogaster (Enjin et al., 2016), is markedly reduced in A. lycopersici compared to most insects and M. occidentalis (Hoy et al., 2016). However, the numbers are similar to those in T. urticae (each has four putative IRs with strong bootstrap support), including homologues of the highly conserved IR25a and IR93a receptors (Figure 4—figure supplement 8). Interestingly, A. lycopersici may have as few as six ionotropic glutamate receptors (iGluRs), as compared to 14 in T. urticae (Figure 4—figure supplement 8); proteins in this family are related to IRs, but have ultra-conserved roles in synaptic transmission in animals (Benton et al., 2009).

Finally, we found both expansions and contractions of the TRP family (Figure 4—figure supplement 9). Like the other sequenced herbivorous mite, T. urticae, no orthologue of TRPA1 was located, but orthologues for TRPgamma, NopmC, and TRPML are present, with three copies of NopmC as compared to T. urticae's two. Unlike T. urticae, members of the TRPP and TRPM clades were completely absent in the russet mite, but strikingly, two putative members of the TRPV clade (Inactive and Nanchung), previously thought to be lost in mites and ticks (Peng et al., 2015; Regier et al., 2010), appear to be present.

Among two vertebrates, one nematode and the 15 arthropod species we analyzed, A. lycopersici has the lowest number (364) of transcription factor (TF) genes (Supplementary file 1 — ‘Table S11’ Tab). Nevertheless, when accounting for the total number of genes by species, the TF fraction in A. lycopersici (3.55%) is higher than that of T. urticae (2.98%), and is within the range reported for metazoan animals (4.7% ±1.4, Charoensawan et al., 2010). However, a lower number of the PFAM TF domains Zinc finger (zf-C2H2 and zf-CCHC), Forkhead, Homeobox, Hormone (nuclear) receptor, HLH, bZIP_2 and T-box were found in A. lycopersici compared to all other species included in our analysis (Supplementary file 1 — ‘Table S11’ Tab). In addition, A. lycopersici orthologues of the Hairy Orange protein family (hey, cwo and deadpan) have lost the Hairy Orange domain (Figure 4—figure supplement 10), while an orthologue of D. melanogaster SoxNeuro could not be identified in A. lycopersici despite being present in the spider and Acari genomes examined (Figure 4—figure supplement 11). Among nuclear receptors (NRs), we identified eight canonical NRs in the A. lycopersici genome (E78, HR3, EcR, two RXRs, ERR, FTZ-F1, HR96) that contained both a DNA-binding domain (DBD) and a ligand-binding domain (LBD). However, no homologues of the evolutionary conserved NRs HNF4, HR39, HR78, and HR83 (Bodofsky et al., 2017; Bonneton and Laudet, 2012), nor a homologue of the T. urticae Photoreceptor-specific NR (PNR), were detected in the A. lycopersici genome, even though HR78, HNF4, and PNR are present in D. pteronyssinus (Supplementary file 1 — ‘Table S12.1’ Tab and ‘Table S12.2’ Tab). Further, for six nuclear receptors (E75, DSF, HR4, HR38, HR51, and SVP) that are evolutionary conserved across arthropods and normally have a canonical (DBD+LBD) structure (Fahrbach et al., 2012; Grbić et al., 2011; Hwang et al., 2014; Litoff et al., 2014), an LBD was not predicted for the respective A. lycopersici homologues. LBDs for all of these except HR4 were predicted for both the D. pteronyssinus and T. urticae homologues (Supplementary file 1 — ‘Table S12.1’ Tab and ‘Table S12.2’ Tab, Figure 4—figure supplements 12–17).

The basic helix-loop-helix (bHLH) gene family is an ancient family found in fungi, plants, and animals, and members of this family are essential both for organisms to respond to environmental factors, as well as for cellular differentiation during development (Skinner et al., 2010). The D. melanogaster achaete and scute bHLH genes play crucial roles in bristle development (García-Bellido and de Celis, 2009). Within the bHLH family group we found that T. urticae, M. occidentalis and I. scapularis have five bHLH proteins with an achaete-scute InterPro domain (IPR015660), while only three were found in both D. pteronyssinus (g4111.t1, g7028.t1 and g6164.t1) and A. lycopersici (aculy01g18470, aculy01g18540 and aculy02g28230).

A number of other specific transcription factors that are highly conserved among most arthropods are also absent from the A. lycopersici genome. For A. lycopersici, we were unable to identify proboscipedia, a member of the Hox gene family. Members of this family (labial, proboscipedia, Hox3/zen, Deformed, Sex combs reduced, fushi tarazu, Antennapedia, Ultrabithorax, abdominal-A, and Abdominal-B) encode homeodomain transcription factors and act to determine the identity of segments along the anterior–posterior axis in arthropods (Hughes and Kaufman, 2002). Proboscipedia is present in all chelicerate genomes (horseshoe crab, scorpions, spiders, mites and ticks) for which Hox genes have been analyzed (Figure 5, Supplementary file 1 — ‘Table S13.1’ Tab and ‘Table S13.2’ Tab, Supplementary file 6; Di et al., 2015; Hoy et al., 2016; Kenny et al., 2016; Schwager et al., 2017), and is believed to be ancestral to all arthropods (Pace et al., 2016). Of particular note, proboscipedia is located in close proximity (<35 kb) of labial in Acariformes, but in Aculops labial was the only Hox gene that was present on scaffold 2 (Supplementary file 1 — ‘Table S14’ Tab). Furthermore, A. lycopersici lacks a homologue of the T-box encoding gene org-1 (Figure 4—figure supplement 18), which in D. melanogaster plays a pivotal role in diversification of circular visceral muscle (Schaub and Frasch, 2013). Finally, we also mined the A. lycopersici genome for transcription factors and other genes involved in circadian rhythm (so-called ‘clock’ genes) (Supplementary file 1 — ‘Table S15’ Tab). Orthologues of the helix-loop-helix TFs cycle, Clock and tango and the bZIP TF vrille were identified in the A. lycopersici genome. However, we did not identify period and timeless, known negative regulators of Clock and cycle (Lee et al., 1999; Peschel and Helfrich-Förster, 2011). Other circadian regulators, like the circadian photoreceptor cryptochrome and the bZIP TF PAR-domain protein 1ε, were also not identified, even though these are present in T. urticae (Hoy et al., 2016).

We identified 18 putatively intact horizontal gene transfer (HGT) candidate genes (Supplementary file 1 — ‘Table S16’ Tab), and performed subsequent phylogenetic analyses that suggested that nine were acquired from a foreign source. Seven of these genes code for UDP-glycosyltransferases (UGTs), members of which have well documented roles in xenobiotic detoxification (Snoeck et al., 2019). Phylogenetic inference with all T. urticae¸ D. pteronyssinus and A. lycopersici UGTs (80, 27, and 7, respectively) indicated that the seven UGTs in the tomato russet mite genome were the result of a lineage-specific expansion (Figure 4—figure supplement 19). Although we did not observe a clear phylogenetic signature of HGT (Wybouw et al., 2016), our phylogenetic reconstruction is consistent with previous studies which indicated that, prior to the formation of the Acariformes lineage, an ancestral mite species laterally acquired a UGT gene copy from a bacterial source (Ahn et al., 2014; Wybouw et al., 2018).

Two intact genes of bacterial origin (aculy01g38350 and aculy04g02470) were also identified in the tomato russet mite genome that are predicted to code for enzymes in the microbial and plant pantothenate biosynthesis pathway (an apparent duplicate of aculy01g38350 was also uncovered, but the coding sequence was disrupted, and it lacked expression, suggesting it is a pseudogene) (Figure 6). PCR amplification linked both laterally acquired genes with either neighboring intron-containing genes (aculy01g38350) or conserved eukaryotic genes (aculy04g02470 is located next to aculy04g02480, which encodes a Gtr1/RagA protein); in addition, an aculy01g38350 transcript (Illumina contig 1934) had a polyA tail, suggestive of eukaryotic transcription (Figure 6—figure supplement 1). Pantothenate, or vitamin B5, is a life-essential compound, and whereas plants and bacteria are able to synthesize this compound de novo, animals rely on dietary uptake. Genes for pantothenate synthesis are present in tetranychid mites, and genomic and phylogenetic approaches have pointed to an ancient HGT event prior to speciation within the Tetranychidae family for both genes. Constrained tree tests rejected the topology where ketopantoate hydroxymethyltransferase of A. lycopersici was the sister lineage to the group of spider mite biosynthetic proteins, but not for pantoate β-alanine ligase, suggesting that A. lycopersici acquired the ketopantoate hydroxymethyltransferase gene from a different bacterial donor species (Figure 6, Approximately Unbiased tests, p-value cut-off of 0.01).

In T. urticae the acquisition of pantothenate biosynthetic genes is accompanied by the horizontal gene transfer of two methylenetetrahydrofolate dehydrogenases (MTHFDs), enzymes of the folate pathway and connected to the pantothenate biosynthesis pathway (Wybouw et al., 2018). Although such a HGT was not detected in A. lycopersici, an expansion of MTHFDs was detected compared to other mite species (OG0000706 in Supplementary file 1 — ‘Table S7’ Tab).

Small molecules or proteins produced in salivary glands are one mechanism by which arthropod herbivores can manipulate the defenses of their host plants. As A. lycopersici is able to potently suppress tomato defenses (Glas et al., 2014; Schimmel et al., 2018), we predicted its secretome, and found that 612 of the 10,263 annotated A. lycopersici proteins (6%) are putatively secreted (Supplementary file 1 — ‘Table S17’ Tab). Only one of the more than 600 secreted A. lycopersici proteins (aculy02g17370, a glycosyl hydrolase, family 13, IPR013780) had a best BLASTp hit with a T. urticae protein that was previously identified in T. urticae saliva using an LC-MS/MS (Jonckheere et al., 2016). More than half (351) of these proteins were absent in orthogroups in non-herbivorous arthropod species, and are less than 350 amino acids in length. Only 15 of these 351 proteins belonged to an orthogroup with more than one member in A. lycopersici (OG0006384, OG0009325, OG0009954 and OG0010904). Among these, OG0009325 contains three short A. lycopersici proteins <90 amino acids in length (aculy01g11450, aculy01g12600, and aculy01g12690). Of note, the gene encoding the single T. urticae representative in this group, tetur24g01070, was previously found to be overexpressed in the T. urticae salivary gland region (Jonckheere et al., 2016). OG0006384, on the other hand, contains cysteine peptidases (Peptidase C1A, papain C-terminal domain; InterPro IPR000668), which are enzymes reported to have key roles in plant-pathogen/pest interactions (Shindo and Van der Hoorn, 2008), and for which two lineage-specific expansions are present in A. lycopersici (Figure 4—figure supplement 20).

We also characterized components of small RNA pathways that might be of potential relevance for agricultural control methods. The A. lycopersici genome harbors highly conserved miRNA sequences, such as let-7, miR-1, and miR-9a (Supplementary file 1 — ‘Table S18’ Tab). However, in contrast to T. urticae, a clear A. lycopersici homologue of Exportin-5, a dsRNA-binding protein mediating nuclear transport of pre-miRNAs (Bohnsack et al., 2004; Kim, 2005), is lacking, suggesting a deviating miRNA pathway in A. lycopersici. In line with the latter hypothesis, we could not identify an A. lycopersici homologue of Staufen, while this gene is present in T. urticae (Grbić et al., 2011; Supplementary file 1 — ‘Table S19’ Tab) and other arachnids (OrthoDb v 9.1, group EOG091G07A0 and EOG090Z04UZ, respectively) and was shown to negatively modulate miRNA activity in the nematode C. elegans (Ren et al., 2016).

The A. lycopersici genome contains, in line with T. urticae, clear homologues of Dicer-1, Loquacious, Drosha and Pasha and an expansion of the AGO1 and PIWI/AGO3 subfamilies. Of note, we found one A. lycopersici protein (aculy02g00240) that was highly homologous to the T. castaneum Dicer-1 enzyme (bitscore of 294) and that contained both an RNA-binding domain (PAZ-domain, cl00301) and the RNAse III domain (cd00593) while two A. lycopersici proteins (aculy02g04810 and aculy02g19970) showed reciprocal BLASTp hits with T. castaneum Dicer-2 and Dicer-1, respectively, but were relatively short (about 500 amino acids (aa) compared to 1726 aa for aculy02g00240) and only contained the RNAse III domain. However, the genes encoding these proteins are located next to a sequencing gap in the current assembly and it could be that gene-models for these Dicer-like enzymes are not complete. Similar to T. urticae, we could not identify clear homologues of R2D2 and AGO2 (Grbić et al., 2011; Supplementary file 1 — ‘Table S19’ Tab), suggesting that the siRNA pathway is either absent or non-canonical in both mite species (Okamura et al., 2011).

Further, important players in the PIWI-interacting RNA (piRNA) pathway (Iwasaki et al., 2015) were identified in the A. lycopersici genome (PIWI/AGO3, Zucchini, Armitage, Maelstrom and SoYb; Supplementary file 1 — ‘Table S19’ Tab), while homologues of Armitage and Zucchini could not be identified in T. urticae, which is in line with the recently suggested non-canonical piRNA pathway in T. urticae (Supplementary file 1 — ‘Table S19’ Tab, Huang et al., 2014; Mondal et al., 2018b).

Finally, RNA-dependent polymerases are known to be essential for the amplification of the RNA silencing effect (systemic RNAi) in C. elegans and some plants (Tomoyasu et al., 2008). Genes encoding these enzymes are absent in insect genomes while 1 to 5 have been reported in Acari genomes (Grbić et al., 2011; Hoy et al., 2016; Joga et al., 2016; Mondal et al., 2018a; Zong et al., 2009). Surprisingly, we could not identify RNA-dependent polymerase genes in the A. lycopersici genome (Supplementary file 1 — ‘Table S19’ Tab), which might indicate that these genes have been lost since the divergence of Eriophyoidea from other acariform lineages. However, as systemic RNAi does seem to occur in some insect orders, for example, Coleoptera (Joga et al., 2016), we cannot exclude that systemic RNAi might also occur in A. lycopersici.

At only 32.5 Mb, the A. lycopersici genome is the smallest sequenced arthropod genome to date. In contrast to its closest sequenced relatives, the majority of genes lack introns, few repetitive sequences are present, and many genes conserved in most animals are absent. Compared to its larger relatives, the simplification of A. lycopersici’s body plan, and that of eriophyoid mites more generally, is reminiscent of that observed in other microarthropods (Maderspacher, 2016). The compressed genome architecture of A. lycopersici is in line with genome streamlining concepts (Hessen et al., 2010a; Hessen et al., 2010b), some of which speculate that maintaining a high growth rate in nutritionally limited environments (in this study the plant epidermis) may be a driver for the evolution of compact genomes. Further, the extreme reduction of several environmental response gene families aligns with predictions that follow from ecological specialization theories (Devictor et al., 2010; Futuyma and Moreno, 1988; Laland et al., 2016) since the mite’s suppression of plant defenses may allow for such families to minimize during the course of its evolution. Finally, this first eriophyoid genome provides a resource for methods of early detection of mite infestations using molecular markers, and its reduced complement of defense genes – a common source of pesticide resistance – may also reveal novel Achilles’ heels for the control of A. lycopersici. But foremost, this genome is a milestone for accelerating our understanding of the evolutionary forces underpinning metazoan life at the limits of small physical and genome size.

A. lycopersici individuals were reared in insect cages (BugDorm-44590DH, Bug Dorm Store, MegaView Science, Taichung, Taiwan) in a walk-in growth chamber on tomato plants (Solanum lycopersicum, cv. Castlemart) that were between 3 and 6 weeks old. The climate room was set to day/night temperatures of 27°C/25°C, a 16/8 hr light/dark regime and 60% relative humidity. Harvesting of A. lycopersici mites was performed by detaching highly infested tomato leaflets and placing them in 1.5 mL Eppendorf tubes. Eppendorf tubes were filled with water and mites (adults, juveniles and eggs) were washed off by rinsing and briefly vortexing the tubes. The tubes were then centrifuged (13,000 rpm for 2 min), after which bulk tomato tissue was removed and water was pipetted away. Contamination from tomato tissue was limited to small amounts (less than ~5%) of material consisting primarily of tomato trichomes. Resulting ‘pellets’ of russet mites were frozen in liquid nitrogen and stored at −80°C until DNA was extracted.

DNA was extracted using a modified version of the CTAB method (Doyle and Doyle, 1987). Sixty µg of DNA dissolved in TE buffer was sent to Eurofins MWG Operon (Ebersberg, Germany) for sequencing. Sequencing reads were produced with the standard Roche/454 sequencing protocol on the GS FLX system running Data Analysis Software Modules version 2.3. Three different libraries were prepared and sequenced in accordance with the recommendations of Roche/454: random primed shotgun, 8 kb paired-end, and 20 kb paired-end. From the shotgun library the mean length was 503 bp, while for the 8 kb and 20 kb libraries the mean lengths were 366 bp and 359 bp, respectively. Sequencing reads were trimmed to remove adapters and low-quality bases, as well as to split each paired-end read into a forward and reverse pair; this yielded a total of 1,854,028 shotgun reads, 1,076,303 reads from the 8 kb library, and 1,274,414 reads from the 20 kb library. Contigs were assembled by the in-house pipeline of Eurofins MWG Operon (Ebersberg, Germany) based on Newbler 2.6 (Margulies et al., 2005). Following scaffolding and filtering for plant (tomato), prokaryotic, and adaptor sequences, a reference for the nuclear genome was generated that consisted of seven scaffolds (scaffolds 1, 2, 3, 4, 5, 11, and 17) with a total length of 32.53 Mb (the Newbler ‘peakDepth’, or coverage, for the assembly was 38). An additional scaffold (scaffold 6) of length 13.5 kb consisted of the mitochondrial genome.

A k-mer size estimate of the A. lycopersici genome was performed using the genomic 454 sequence reads and Jellyfish 2.2.6 (Marçais and Kingsford, 2011). Following the recommendations of T. Nishiyama (http://koke.asrc.kanazawa-u.ac.jp/HOWTO/kmer-genomesize.html), genome size was estimated by running Jellyfish (Marçais and Kingsford, 2011) with the following settings ‘-t 24 iC -s 20M’ for all odd k-mer values from 17 to 31, with averaging of the results provided from the eight different estimates. Completeness of the genome was also assessed using CEGMA 2.5 (Parra et al., 2007) as well as BUSCO v3 (Simão et al., 2015), as well as with an alignment of the A. lycopersici Illumina-based transcriptome assembly to the genomic scaffolds (see below, and Results section).

Mixed developmental stages (adults, juveniles, and eggs) were collected from tomato leaflets as was done for DNA preparation. Similar to DNA extraction, small amounts of tomato trichome contamination were evident, but at low levels. RNA was extracted using a Qiagen RNeasy kit (Qiagen, Hilden, Germany) according to the manufacturer’s instructions. Forty-five µg of RNA was provided to Eurofins MWG Operon for library preparation according to standard Roche protocols. Following poly(A) selection and strand-specific cDNA library preparation, the library was analyzed on a Shimadzu MultiNA microchip electrophoresis system (Shimadzu, Kyoto, Japan) to verify that the gel size selection was in the range of 500–800 bp. A total of 1,370,892 sequencing reads were collected using a Roche GS FLX system employing the Titanium series chemistry. After trimming of cDNA reads to remove low quality reads and adapter sequences, the remaining 1,370,005 reads were assembled using MIRA (Chevreux et al., 2004).

RNA was extracted from eight A. lycopersici pools using the Qiagen RNeasy purification kit (Qiagen, Hilden, Germany) with the following adaptations: Step 3: 50 µl of RNEasy lysis buffer (RLT) + ß -mercaptoethanol were added to the mite pool in a 1.5 mL tube, followed by 1–2 min of cell lysis performed by twisting and turning a 1.5 mL-tube-pestle. Three hundred µl of RLT + -mercaptoethanol was then used to rinse the pestle; Step 11: RNA was eluted in 30 µl RNAse-free water and stored on ice. All samples were stored at −20°C. Strand-specific paired-end RNA library preparation and sequencing were carried out by the Centro Nacional de Análisis Genómico (Barcelona, Spain) to yield a total of 86.6 million 101 bp read pairs.

To construct a transcriptome assembly from the Illumina RNA-seq reads, the reads were first aligned to the A. lycopersici reference genome sequence using STAR 2.5.2b (Dobin et al., 2013) with the following settings: twopassMode Basic, sjdbOverhang 100, and alignIntronMax 20000. Reads that did not align to the reference were subsequently aligned against the tomato genome release SL 2.50 (Tomato Genome Consortium et al., 2012) to filter out contamination from the host plant with the same settings used to align to the mite genome except for alignIntronMax, which remained unspecified. The reads that did not align to the tomato genome were pooled with the reads that aligned to the A. lycopersici genome and imported into CLC Genomics Workbench 9.0.1 (https://www.qiagenbioinformatics.com/), where they were trimmed using the default parameters (quality score limit 0.05 and a maximum of two ambiguous nucleotides) before being assembled with the default settings and a minimum contig length of 200. The resulting 13,428 transcript sequences were aligned back to the A. lycopersici genome assembly using BLAST 2.5.0+ (Camacho et al., 2009) to provide a measure of the genome completeness for transcribed regions. Of the 243 transcripts that did not align, 23 had no hits in any database, and 108, 84 and 20 appeared to be from bacterial, plant and fungal sources, respectively. Only eight had homology to arthropod sequences present in the NCBI NR, NT, Other Genomic, RefSeq Genomic, RefSeq RNA, Representative Genomes, and WGS databases (downloaded January 9, 2017).

A first-pass annotation was produced using EuGene (Schiex et al., 2001) specifically trained for the studied genome using the 454 transcript read data as a guide. As a consequence of the close proximity of adjacent genes (see Results and Table 1), we observed that transcript contigs often merged adjacent genes, creating apparent chimeric genes. To circumvent this issue, only junctions spanning introns as assessed from the aligned 454 data were kept after mapping. Besides using transcript data, protein homology to the invertebrate section from RefSeq, curated proteins from SWISSprot and the proteome from T. urticae were used.

Subsequently, the annotation was revised in several ways. The deep dataset of Illumina RNA-seq reads was aligned to the genome using the default settings of Bowtie 2.2.3 (Langmead and Salzberg, 2012)/TopHat 2.0.12 (Kim et al., 2013), as well as STAR 2.5.2b (Dobin et al., 2013) with the parameters described previously. Transcripts from the CLC transcriptome assembly were also located on the genome using BLAT 36 (Kent, 2002). Then, Cufflinks 2.2.1 (Trapnell et al., 2013) and TransDecoder (Release 20140704) (Haas et al., 2013) were used to identify additional ORFs of over 300 bp in length that had not been detected by EuGene. Resulting gene models were then added where supported by the strand-specific RNA-seq reads and/or transcript alignments. The compact nature of the A. lycopersici genome, coupled with the finding that most genes were intronless (Table 1), made it feasible to then manually inspect all gene models against the aligned Illumina RNA-seq read data. This inspection step was performed using the Integrative Genomics Viewer (Robinson et al., 2011), which allowed simultaneous display of gene models and RNA-seq read alignments. Manual adjustments to gene models, where required, were performed using GenomeView N29 (Abeel et al., 2012). Additionally, members of specific gene families were expertly annotated as described in the section ‘Comparative analyses with specific gene families’, with resulting adjustments also incorporated in the final annotation. GenomeTools 1.5.10 (Gremme et al., 2013) was used to sort, correct phase information, and validate the resulting GFF3.

Coding gene numbers and the percentages of intronless genes were calculated with the ‘stat -exonnumberdistri’ command of the GenomeTools 1.5.6 package (Gremme et al., 2013) using the respective GFF3 annotation files as input (Table 1). Where multiple isoforms were present for a gene, only the longest isoform was used for this and subsequent analyses. Regions of the respective genomes were then classified as coding, intergenic or intronic by parsing the location of coding sequences (CDS) from the respective GFF3 annotation files; due to the unreliability of untranslated sequence prediction or their complete absence in some annotations, these regions were not considered. In instances where CDS sequences overlapped, their coordinates were merged so that no region of the genome would be counted multiple times. Regions of the genome between the start and end of the CDS sequences of adjacent genes were classified as intergenic, while regions of the genome within genes that did not fall into CDS coordinate blocks were classified as intronic (in instances where genes were located within the introns of other genes, the CDS sequences of the genes within the introns were classified as coding, with the remaining portion counted as intronic).

The consensus of the repeated DNA (≥2 copies) in the genome was constructed by employing RepeatScout (v.1.0.5) (Price et al., 2005). The repeats that were ≥90% identical with a minimum overlap of 40 bp were assembled using CAP3 (Huang and Madan, 1999). Gene families were identified based on homology with cellular genes by employing tBLASTx 2.2.28+ (Altschul et al., 1997) searches against the Refseq mRNA database at NCBI and BLASTn 2.2.28+ (Altschul et al., 1997) searches against the annotated genes in the A. lycopersici genome. All candidate gene families were filtered upon manual verification. The remaining repeats were classified by REPCLASS (Feschotte et al., 2009) and RepeatMasker (Smit et al., 2013) protein searches (http://www.repeatmasker.org/cgi-bin/RepeatProteinMaskRequest). The repeats that were classified based on the structure or TSD module of REPCLASS were manually verified. The criteria of requiring at least one defined end were used to classify a repeat as a TE. To identify if the elements had at least one defined end, the unclassified repeats (≥65 bp) were aligned with the respective copies with extended flanking sequences using MUSCLE (Edgar, 2004). Repeats were classified and full-length copies were extracted when possible. To identify low copy non-LTR retrotransposons, the non-LTR proteins from the related mite T. urticae were used as queries in homology-based tBLASTn 2.2.25+ (Altschul et al., 1997) searches against the A. lycopersici genome. To identify the genomic coverage, the curated repeat library was used to mask the genome using RepeatMasker (v 4.0.5) (Smit et al., 2013). The final RepeatMasker output was parsed using parseRM.pl (Kapusta et al., 2017; Kapusta, 2017) to identify the contribution of TEs (Figure 2—figure supplement 1, Supplementary file 1 — ‘Table S1’ Tab). Last, a gene and TE density plot was constructed using karyoploteR version 1.14.0 (Gel and Serra, 2017) and the GFF3 annotation file of the A. lycopersici genome (Table 1—source data 1) and the RepeatMasker output (Supplementary file 1 — ‘Table S2’ Tab), respectively.

The longest protein isoforms for the following organisms were extracted for orthologue identification: A. lycopersici (current genome), Anopheles gambiae AgamP4.7 (Holt et al., 2002), Bombyx mori ASM15162 (Ensembl release 37) (Mita et al., 2004), Caenorhabditis elegans Wormbase release WS261 (The C. elegans The C. elegans Sequencing Consortium, 1998), Centruroides sculpuratus CEXE 0.5.3 (Schwager et al., 2017), Danio rerio GRCz10 (Ensembl release 89) (Howe et al., 2013), Daphnia pulex PA42 3.0 (Ye et al., 2017), Dermatophagoides pteronyssinus (ASM190122v2) (Waldron et al., 2017), Drosophila melanogaster Flybase release 6.16 (Adams et al., 2000; Gramates et al., 2017), Homo sapiens GRCh38.p10 (Ensembl release 89) (Lander et al., 2001; Venter et al., 2001), Ixodes scapularis (IscaW1.5) (Gulia-Nuss et al., 2016), Limulus polyphemus 2.1.2 (Simpson et al., 2017), Metaseiulus occidentalis 1.0 (GNOMON release) (Hoy et al., 2016), Parasteatoda tepidariorum 1.0 (Schwager et al., 2017), Pediculus humanus PhumU2 (Ensembl release 36) (Kirkness et al., 2010), Strigamia maritima Smar1 (Ensembl release 36) (Chipman et al., 2014), T. urticae (ORCAE August 11, 2016 release) (Grbić et al., 2011), and Tribolium castaneum Tcas5.2 (Ensembl release 36) (Richards et al., 2008). The identification of orthologous protein sequences was performed with OrthoFinder 1.1.8 (Emms and Kelly, 2015) using BLAST 2.6.0+.

We found 147 single-copy orthologues across all species that we then aligned using MAFFT 7.305b (Katoh and Standley, 2013) with ‘genafpair’ and ‘maxiterate 1000’; a concatenation of the alignments for the 147 orthologues was then generated prior to trimming with trimAl 1.4.rev15 (Capella-Gutiérrez et al., 2009) using the ‘strictplus’ option. The trimmed sequences were used for a phylogenetic reconstruction with RAxML 8.2.12 (Stamatakis, 2014) using the LG+G+F model as identified for phylogenetic reconstruction by ProtTest 3.4.2 (Darriba et al., 2011) according to the Akaike Information Criterion, and a total of 1000 rapid bootstrap replicates (‘-f a -x 12345’ option). Although the ‘estimate proportion of invariable sites (+I)’ was also recommended by ProtTest, the developer of RAxML v8, on page 59 of the RAxML v8.2.X manual, cautions against using this option, and this and all subsequent optimal models for reconstructions with RAxML were adjusted to adhere to this developer recommendation.

Orthologous protein clusters were selected for intron analysis on the basis of the following criteria: the cluster had to have at least one orthologue from A. lycopersici, orthologous protein sequences from at least 14 other species had to be present, and no species could have more than three orthologous proteins in the cluster; when multiple orthologues for a single species were present, only the longest one was retained. The sequences in these clusters were aligned using MAFFT 7.305b (Katoh and Standley, 2013) with the settings previously described. GNU Parallel (Tang, 2011) was used to align multiple clusters at once. Custom Python scripts using BioPython 1.70 (Chapman and Chang, 2000) and the BCBio GFF parser (Chapman, 2016) were used to parse and append intron position information to the FASTA sequence identifier line as required by Malin (Csurös, 2008). The 2371 clusters that met the requisite criteria, along with the tree built from the 147 single-copy orthologues, were used in the Malin analysis (Csurös, 2008). Intron positions for gain/loss analysis were selected from those that were considered unambiguous in A. lycopersici and at least 11 other species, with five amino acids present on either side of the intron position (a Malin criteria to reduce the possibility of incorrect inference resulting from misalignments).

To investigate the consequence of intron losses in A. lycopersici on predicted protein sequences, which can shed light on underlying mechanisms of loss (see Discussion), a subset of orthogroups was selected for which sequences for each of A. lycopersici, D. pteronyssinus, T. urticae, M. occidentalis, B. mori and D. melanogaster were present as single copies (1216 in total); apart from A. lycopersici, the five other species were selected because of their close phylogenetic position to A. lycopersici (Figure 3), and/or because they have high-quality genomes and annotations. The protein sequences for the six species for each of these orthogroups were aligned using MAFFT 7.407 with the settings previously described, and a table of intron sites for these orthogroups was generated in Malin using the following settings: Minimum nongap positions: 0 (On both sides); Minimum unambiguous characters at a site: 1; There must be at least one unambiguous character in the following clades: All unselected. From this table, intron positions that were present and had the same phase in all arthropods except A. lycopersici (indicating a high degree of conservation), and for which Malin identified a missing intron in a region of unambiguous alignment for A. lycopersici sequences, were manually examined across all protein sequence alignments to assess if intron loss events in the respective genes introduced gains or losses of residues in the encoded products. The classification results for these sites (100 in total among 80 orthogroups), are included in Supplementary file 1 — ‘Table S3’ Tab; the sequence alignments and annotations of intron positions for each orthogroup are given in Supplementary file 3.

The OrthoFinder analysis (see section ‘Analysis of intronic features’) generated 86,686 orthologous groups (OGs) in total, of which 13,817 contained more than one protein (Supplementary file 1 — ‘Table S7’ Tab). InterProScan 5.25–64.0 (Quevillon et al., 2005) was run to assign domains to each of the proteins in all 18 species, and the domain information was subsequently assigned to the OrthoFinder OGs using the KinFin software (Laetsch and Blaxter, 2017) and an associated Python script (functional_annotation_of_clusters.py with the options: ‘–p 0.3 and –x 0.3’). Two different strategies were used to identify contracted and/or expanded gene families in A. lycopersici. First, we used the CAFE software to detect contracted/expanded orthologous groups (orthogroups, OGs) among 18 metazoan species, while the second strategy was focused on OG expansions within the acariform mites, A. lycopersici, D. pteronyssinus and T. urticae using an arbitrary rule. OrthoFinder 1.1.8 (Emms and Kelly, 2015) with BLAST 2.6.0+ was used to identify OGs among the proteomes of 18 metazoan species (see ‘Analysis of intronic features’ for proteome versions that were used as input for OrthoFinder).

To maximize the probability of achieving convergence in the maximum likelihood analysis performed in CAFE, OGs were processed to remove OGs present in only a few species and were subsequently divided into OGs having <100 gene copies in any species (‘small’ OGs) and orthogroups having one or more species with ≥100 gene copies (‘large’ OGs); see ‘Known Limitations’ section in CAFE 4.0 Manual of March 14, 2017 and section 2.2.4 of the CAFE 4.0 tutorial online at https://iu.app.box.com/v/cafetutorial-pdf, and also Casola and Koralewski, 2018. We retained 6,496 OGs that occurred in no less than 10 out of 18 species consisting of 6,467 ‘small’ OGs and 29 ‘large’ OGs. Together with an ultrametric species tree the ‘small’ OG dataset was used as input in CAFE to estimate the birth/death parameter λ (the probability that a gene will be gained or lost) and to identify rapidly evolving OGs (p-value threshold of 0.05). The estimated λ (0.00055594301461) was then used to identify rapidly evolving OGs in a CAFE analysis with ‘large’ OGs and using the same p-value threshold and ultrametric species tree as in the CAFE analysis with ‘small’ OGs. The ultrametric species tree used as input in both CAFE analyses was obtained by using the species tree generated for the Malin intron analysis, subsequently rooting this tree using vertebrates as outgroup, and converting this rooted tree into an ultrametric tree using the convert_to_ultrametric() command in the Tree package of the ETE toolkit (ete 3.0.0b35) (Huerta-Cepas et al., 2016). Next, branch lengths of the ultrametric tree were scaled to time units using the software treePL (Smith and O'Meara, 2012) with the following options: 'smooth = 0.01, numsites = 41107 (number of sites in the alignment used for the Malin analysis), thorough, opt = 4, moredetailad, optad = 2, optcvad = 2, moredetailcvad’ and using seven calibration timepoints: the divergence time between Eriophyoidea and Sarcoptiformes (352–410 MYA), Sarcoptiformes and Trombidiformes (410–421 MYA) and Mesostigmata and Ixodida (283–418 MYA) as derived from Xue et al., 2017, and the divergence time between D. melanogaster and A. gambiae (211–335 MYA), Scorpiones and Araneae (379–410 MYA), Mandibulata and Chelicerata (560–642 MYA) and H. sapiens and D. rerio (425–446 MYA), as obtained from TimeTree (Kumar et al., 2017) on February 20, 2019. The options used in treePL were determined following the ‘Quick run’ guidelines of the treePL wiki (Smith, 2012). The output of the two CAFE analyses (‘small’ and ‘large’ OGs) was summarized using a Python script (cafetutorial_report_analysis.py using the ‘-l’ option and with a p-value cutoff set to 0.05) available at the CAFE tutorial website (https://iu.app.box.com/v/cafetutorial-files/folder/22161236877, accessed February 20, 2019). The tree with OG expansions and contractions was visualized in MEGA 6.0 (Tamura et al., 2013) and edited with Corel Draw software (Corel Draw, Inc); the list of rapidly evolving (expanding or contracting) OGs can be found in Supplementary file 1 — ‘Table S5’ Tab. Rapidly contracting A. lycopersici gene families with more than ten members were analyzed for the percentage of A. lycopersici members showing orthology with the majority of chelicerate species (Supplementary file 1 — ‘Table S6’ Tab). Orthology was determined based on the Orthofinder generated output in the ‘Orthologues_Aculops_lycopersici’ folder. One of the six rapidly contracted A. lycopersici families belonged to the CYP family and was excluded from the analysis, as only few orthology relationships has been observed within this family (Feyereisen, 2011).

Apart from gene families that we identified as expanded in the high-level CAFE analysis, we looked as well for more subtle expansions across acariform mites. Across all orthogroups identified by Orthofinder, we identified eleven orthogroups with (1) A. lycopersici having more than five members and (2) A. lycopersici having twofold more members than the average number in T. urticae and D. pteronyssinus (OG0000024, OG0000271, OG0000546, OG0000706, OG0004829, OG0006109, OG0006384, OG0007553, OG0007554, OG0008410, OG0008412). For two orthogroups (OG0007554, OG00084112), no InterPro domain could be assigned, while OG0000271, OG0000706, OG0004829, OG0006384, OG0007553, and OG0008410 contained proteins with a DnaJ domain (IPR011701), Formate-tetrahydrofolate ligase domain (IPR000559), Acyltransferase 3 domain (IPR002656), a Peptidase C1A domain (IPR000668), Chromo domain (IPR023780) and a Lipase/vitellogenin domain (IPR013818), respectively. The proteins of the three remaining orthogroups (OG0000024, OG0000546, and OG0006109) belonged to the Major facilitator superfamily (MFS, IPR011701 or IPR024989).

For D. melanogaster proteins belonging to orthogroups with (1) members in all included arthropods except A. lycopersici and (2) a maximum of two D. melanogaster members (343 D. melanogaster proteins in total, Supplementary file 1 — ‘Table S8’ Tab), we performed an Over-Representation analysis (ORA) using the WEB-based GEne SeT AnaLysis Toolkit (Liao et al., 2019). An ORA was performed for each GO category (Biological Process, Molecular Function and Cellular Component) using default settings (and ‘genome protein coding’ as reference set) and a Benjamini-Hochberg multiple testing correction (false discovery rate, FDR, of 0.05). In addition, we also identified those orthogroups that contain purported D. melanogaster essential genes, using the list of 427 essential genes provided in the respective study’s first supplementary data table (Aromolaran et al., 2020).

We specifically analyzed genes and gene families associated with herbivory in other animals, as well as those associated with physiological or developmental process related to A. lycopersici’s life history or derived morphology (GSTs, CCEs, CYPs, ABC transporters, MFS proteins, proteases, chemosensory receptors, and transcription factors, including Hox genes). We also characterized genes involved in processes including circadian rhythm, small RNA pathways, and potential regulation of plant defense responses (secreted proteins).

We performed a genomic HGT screen as previously described in Wybouw et al., 2018. Briefly, the A. lycopersici proteome was aligned with metazoan and non-metazoan proteome databases and the bitscores of the best BLASTp hits were recorded. For each protein query, the h-index metric was calculated by subtracting the best metazoan bitscore from the best non-metazoan bitscore. An A. lycopersici gene was designated as a horizontally transferred gene candidate when it exhibited a best non-metazoan bitscore ≥75 and an h-index ≥30. In our screen, we also performed a tBLASTn-search against the tomato russet mite scaffolds using all identified horizontally transferred T. urticae genes as queries. Maximum-likelihood phylogenies were subsequently constructed for all A. lycopersici horizontally transferred gene candidates, except for a putative UGT pseudogene that was located on scaffold 5 between coordinates 140,638 and 140,871. All complete A. lycopersici UGT genes were sent to the UGT Nomenclature Committee to obtain unique UGT gene names (https://prime.vetmed.wsu.edu/resources/udp-glucuronsyltransferase-homepage). For the final phylogenetic reconstruction of the pantothenate biosynthetic genes, homologues of aculy01g38350 (ketopantoate hydroxymethyltransferase, panB) and aculy04g02470 (pantoate β-alanine ligase, panC) were identified by BLASTn and tBLASTn searches (E⁻¹⁰ cut-off) against the nonredundant nucleotide and protein NCBI databases, respectively, and were grouped based on their position in the tree of life (fungi, animals, bacteria, plants, and other). Proteins were selected per group based on manual inspection of the alignments and were combined with homologues as identified by Wybouw et al., 2018. In addition, we also added a panC homologue of the mealybug Ferrisia virgata to the final set of proteins (Husnik and McCutcheon, 2016). For the phylogenetic analysis of UGTs, we added UGT protein sequences from the annotated genome assembly of the house dust mite D. pteronyssinus (Waldron et al., 2017) to our UGT phylogenetic reconstruction. Applying an E-value of E⁻¹⁰ as the cut-off for the alignments, 27 D. pteronyssinus sequences were identified by reciprocal BLASTp-searches between the D. pteronyssinus proteome and the 87 T. urticae and A. lycopersici UGT sequences. Protein sequences were aligned using the online version of MAFFT v7.380 (Katoh et al., 2019) (available at https://mafft.cbrc.jp/alignment/software/) with 1000 iterations and the options ‘E-INS-i’ and ‘reorder’ (see Supplementary file 7). Protein models were selected based on the Akaike Information Criterion using ProtTest 3.4 (Darriba et al., 2011) (panB: LG+G, panC: LG+G, and UGT: LG+G+F). Maximum likelihood analyses were performed using RAxML v8 HPC2-XSEDE (Stamatakis, 2014) on the CIPRES Science Gateway (Miller et al., 2010) with 1000 rapid bootstrap replicates (‘-f a -x 12345’ option). An additional maximum likelihood tree reconstruction with ultrafast bootstrapping (1000 replicates) was performed for the pantothenate biosynthetic proteins using IQ-TREE version 1.6.12 (Hoang et al., 2018; Nguyen et al., 2015). ModelFinder identified LG+I+G4 as the best protein model based on the Bayesian Information Criterion (Kalyaanamoorthy et al., 2017). Constrained tree tests for alternative topologies whereby A. lycopersici is the sister lineage to the spider mite pantothenate biosynthetic proteins were performed using the approximately unbiased test of IQ-TREE version 1.6.12 (10,000 RELL replicates) (Shimodaira, 2002). The random number seed was set at 12345. Last, the physical location of the aculy01g38350 and aculy04g02470 genes in the A. lycopersici genome was examined by PCR amplification. A. lycopersici mites were collected by soaking infested tomato leaves overnight in 40 mL of 70% ethanol. Mites in ethanol were centrifuged at 2000 rpm for 1 min, ethanol was removed, and pelleted mites were ground using liquid nitrogen. One mL of CTAB buffer with 2% beta-mercaptoethanol and 1% proteinase K was added to the ground mites, followed by incubation in a warm water bath at 56°C. Next, samples were washed with 1 ml of choloroform:isoamyl alcohol (21:1) and DNA was precipitated with isopropanol on ice for 1 hr. Primer sequences that successfully amplified genomic regions are listed in Supplementary file 1 — ‘Table S20’ Tab. PCRs were performed using the recommended protocol for Phusion High Fidelity polymerase (Thermo Scientific, The Netherlands) and 1 μL of extracted DNA (50 ng/microL) and 0.2 μM of each primer. PCR conditions for fragment 1 and 3 were 98°C for 30 min, followed by 35 cycles of denaturation at 98°C for 10 s, annealing at 55°C for 30 s, and extension at 72°C for 1 min (fragment 1) or 45 s (fragment 3) followed by a final extension step at 72°C for 5 min. PCR conditions for fragment two were as follows: 98°C for 30 s, 5 cycles of 98°C for 10 s, 65°C for 10 s, 72°C for 60 s, five cycles of 98°C for 10 s, 60°C for 10 s, 72°C for 60 s, and 20 cycles of 98°C for 10 s, 60°C for 10 s, 72°C for 60 s, followed by a final extension step at 72°C for 3 min. Resulting amplicons were Sanger sequenced by Eurofins (Leiden, The Netherlands) using PCR (with ‘PCR’ suffix) and sequencing (with ‘seq’ suffix) primers as indicated in Supplementary file 1 — ‘Table S20’ Tab.