Response to selection for parasitism of a suboptimal, low‐preference host in an aphid parasitoid

Abstract Risks of postintroduction evolution in insects introduced to control invasive pests have been discussed for some time, but little is known about responses to selection or genetic architectures of host adaptation and thus about the likelihood or rapidity of evolutionary shifts. We report here results on the response to selection and genetic architecture of parasitism of a suboptimal, low‐preference host species by an aphid parasitoid, Aphelinus rhamni, a candidate for introduction against the soy bean aphid, Aphis glycines. We selected A. rhamni for increased parasitism of Rhopalsiphum padi by rearing the parasitoid on this aphid for three generations. We measured parasitism of R. padi at generations 2 and 3, and at generation 3, we crossed and backcrossed parasitoids from the populations reared on R. padi with those from populations reared on Aphis glycines and compared parasitism of both R. padi and Aphis glycines among F 1 and backcross females. Aphelinus rhamni responded rapidly to selection for parasitism of R. padi. Selection for R. padi parasitism reduced parasitism of Aphis glycines, the original host of A. rhamni. However, parasitism of R. padi did not increase from generation 2 to generation 3 of selection, suggesting reduced variance available for selection, which was indeed found. We tested the associations between 184 single nucleotide polymorphisms (SNP) and increased parasitism of R. padi and found 28 SNP loci, some of which were associated with increased and others with decreased parasitism of R. padi. We assembled and annotated the A. rhamni genome, mapped all SNP loci to contigs and tested whether genes on contigs with SNP loci associated with parasitism were enriched for candidate genes or gene functions. We identified 80 genes on these contigs that mapped to 1.2 Mb of the 483 Mb genome of A. rhamni but found little enrichment of candidate genes or gene functions.


| INTRODUC TI ON
Invasions by exotic species that become pests are an increasing problem for agriculture (Bradshaw et al., 2016). Biological control by introduction of natural enemies has proved effective at reducing the abundance and impact of such pests (Cock et al., 2016), and in principle, provides a safe, cost-effective, and sustainable alternative to widespread application of insecticides (Naranjo et al., 2015). In spite of notable successes, two arguments have been made against the introduction of natural enemies for biological control. The first is that introduced natural enemies may attack native nontarget species and reduce their abundance (Simberloff & Stiling, 1996). Modern methods of host-specificity screening prior to introduction are designed to ensure that organisms introduced for the control of insect (Hoddle, 2005) and weed (McEvoy, 1996) pests do not attack native nontarget species, and thus, the immediate threat of nontarget impacts is greatly reduced. However, a related concern has not been adequately addressed: the possibility that postintroduction evolution in host specificity may result in introduced organisms shifting to attack native nontarget species (Simberloff & Stiling, 1996). If host use can evolve rapidly, then traditional screening will not guarantee the safety of biological control introductions. The second argument against such introductions is that native natural enemies may switch to attack invading pests. Populations of invasive species freed from their native natural enemies can quickly achieve high densities, which may cause behavioural or evolutionary shifts by native natural enemies to attack the invaders. Because of such switches, control of the invading pest may eventually be achieved without the introduction of non-native biological control agents (Carroll et al., 2005;Cox, 2004;Kruitwagen et al., 2018).
Although these issues have been discussed for some time, responses to selection and genetic architectures of host adaptation have been studied in few systems (however, see Auer et al., 2020;McBride et al., 2014;Oppenheim et al., 2012Oppenheim et al., , 2018, and thus, little is known about the likelihood or rapidity of evolutionary shifts in host adaptation after introductions for biological control. The existing evidence concerning postintroduction evolution in host adaptation is weak and controversial (Marohasy, 1996;Secord & Kareiva, 1996;van Klinken & Edwards, 2002). For parasitoids, much is known about their foraging and host selection behaviour (Godfray, 1994;Hoddle, 2005;Strand & Obrycki, 1996), but the evolutionary stability of parasitoid host adaptation is unclear and has received scant attention (Hopper et al., 2005;Hufbauer & Roderick, 2005). The working hypothesis among biological control researchers is that evolutionary changes in host adaptation are genetically complex and therefore unlikely (Hopper et al., 1993). This hypothesis is largely untested: little is known about the genetic architecture of host adaptation for most insects, let alone about the effects of genetic architecture on evolutionary shifts (Hopper et al., 1993(Hopper et al., , 2005Hufbauer & Roderick, 2005;Oppenheim et al., 2012;Oppenheim & Hopper, 2010). The host ranges of parasitoids may be affected by genes underlying a variety of processes, including the ability of female parasitoids to find insects and to recognize and oviposit in those found, as well as subsequent survival of parasitoid progeny in hosts (Vinson, 1976).
Oviposition may be limited by parasitoid decisions (host acceptance) or insect defences, including behavioural and structural defences (Gross, 1993), as well as ecological defences (like ant tending and plant chemistry). Survival after oviposition (host suitability) depends on adequate nutrition and may be affected by synchrony with host development, physiological suppression, host-plant chemistry, and immune responses (for review, see papers in special issue of Journal of Insect Physiology 44 (9): 701-866). Survival may also be affected by bacterial endosymbionts in hosts (Asplen et al., 2014;Hopper et al., 2018;Oliver et al., 2003Oliver et al., , 2005. From an applied perspective, predicting the likelihood of evolution in host adaptation depends on knowledge of how many genes are involved and how they interact (Oppenheim et al., 2012(Oppenheim et al., , 2018: phenotypes that depend on a few genes that interact additively are far more likely to respond rapidly to selection than ones that depend on many genes that interact epistatically. The importance of genetic architecture for response to selection is supported by the results on both the evolution of insecticide resistance (Hardstone & Scott, 2010) and the breakdown of plant resistance to pest insects and diseases (Harris et al., 2003;Neuhauser et al., 2003).
We report here the results on the response to selection for parasitizing a suboptimal, low-preference host species by an aphid parasitoid, Aphelinus rhamni Hopper and Woolley (Hymenoptera: Aphelinidae), and the genetic architecture of this response.
Aphelinus rhamni is being introduced to control Aphis glycines, which has become a major pest of soya bean in North America (Hopper, Lanier, Rhoades, Hoelmer, et al., 2017). In previous research on the laboratory colony studied here, A. rhamni females parasitized Rhopalosiphum padi (L.) (Hemiptera: Aphididae) three times less frequently than Aphis glycines Matsumura (Hemiptera: Aphididae), and this difference arose from differences in both female behaviour and progeny survival (Hopper, Lanier, Rhoades, Hoelmer, et al., 2017).
In that research, 25-minute observations of A. rhamni females exposed to R. padi versus Aphis glycines showed that the parasitoids approached fewer R. padi (3.5 vs. 6.3 aphids), probed fewer R. padi with their ovipositors (2.7 vs. 5.1 aphids) and laid eggs in fewer R. padi (0.4 vs. 2.6 aphids). While the difference in oviposition was fivefold between R. padi and Aphis glycines, the difference in numbers of adult progeny was 11-fold, suggesting mortality of about 50 per cent for eggs laid in R. padi. However, some A. rhamni females parasitized as many as seven R. padi per day and produced as many as five adult progeny per day, suggesting that there might be genetic variation in use of R. padi. To explore this possibility, we selected A. rhamni for increased preference and performance on R. padi by rearing the parasitoid species on this aphid for three generations. We measured parasitism of R. padi at generations 2 and 3, and at generation 3, we crossed and backcrossed parasitoids from the populations reared on R. padi with those from the populations reared on Aphis glycines and compared parasitism of both R. padi and Aphis glycines among F 1 and backcross females. We determined the relationship between genetic markers and parasitism. Finally, we assembled and annotated the A. rhamni genome and tested whether genes near genetic markers were enriched for candidate genes or classes of gene functions. Our results show a genetically based response to selection on A. rhamni for parasitism of R. padi with a complex architecture. Our results imply that parasitoids introduced for biological control may rapidly evolve to attack nontarget species but that this evolution may be limited. As mentioned above, Aphelinus rhamni is being introduced to control Aphis glycines, a major invasive pest of soya bean, and our results suggest that it may shift to attack other aphid species somewhat.

| Study system
Aphelinus species are important in biological control of pest aphids (Hopper, Lanier, Rhoades, Coutinot, et al., 2017;Hopper, Lanier, Rhoades, Hoelmer, et al., 2017;van den Bosch et al., 1959). Like all Aphelinus species, A. rhamni is a koinobiont (host continues to develop after being parasitized) endoparasitoid of aphids. Aphelinus species are small (about 1 mm long) and are weak fliers (Fauvergue & Hopper, 2009), searching for hosts and mates primarily while walking (Fauvergue et al., 1995). Aphelinus females prefer 2nd-to 4th-instar aphids for oviposition, but will oviposit in all stages (Rohne, 2002). At 20℃, wasps develop from oviposited egg to adult emergence in about three weeks. During their third instar, Aphelinus larvae kill their hosts, but leave the host exoskeleton intact, causing it to harden and turn black in a process called mummification (Christiansen-Weniger, 1994), and adults emerge about one week after pupation. Aphelinus rhamni females 1-2 days old carry a mean of 15 mature eggs (Hopper, Lanier, Rhoades, Hoelmer, et al., 2017), but females can produce more eggs daily and so could parasitize as many as 200 aphids during a two-week lifetime (unpublished data). Regiella with PCR and primers specific to these bacteria, but the only endosymbiont found was Arsenophonus in A. glycines (unpublished data), which in other research has been found not to affect parasitism (Wulff et al., 2013). Aphids were reared on their host plants in plant growth rooms at ~20℃, 50%-70% relative humidity and 16:8-h (L:D) photoperiod. Vouchers for these populations are stored at −20℃ in 100% molecular-grade ethanol at BIIRU.
The material was initially screened for hyperparasitoids and pathogens, and a culture was established with seven female and seven male adults for a total of 21 haplotypes, although a limited sample of this number of haplotypes is likely to have captured much of the genetic variation in fitness components (Roush & Hopper, 1995). To maintain genetic variation (Hopper et al., 1993;Roush & Hopper, 1995), the culture was split into four subcultures after one genera- When the response to selection experiment was started, the parasitoids had been in culture 5.5 years or 90 generations. Vouchers for the selection and control populations of A. rhamni are maintained at −20℃ in 100% molecular-grade ethanol at BIIRU.

| Selection regime
To select Aphelinus rhamni for increased preference and performance on R. padi, we put ~200 adult parasitoids into each of three cages (10 cm diameter by 22 cm tall) enclosing the foliage of potted barley infested with several thousand R. padi. For three generations, we transferred ~200 adult parasitoids from each cage to a new cage with aphid-infested barley. Thus, we produced three populations of parasitoids that were exposed to hard selection for three generations: genotypes of females that did not lay eggs in R. padi and genotypes of progeny that did not survive in R. padi would not be represented in the next generation.

| Crosses
After three generations of selection, we crossed females from the control populations (i.e., reared on Aphis glycines) with males from the selection populations (i.e., reared on R. padi) to produce F 1 females, which we reared on Aphis glycines to avoid conditioning or selection for preference/performance on R. padi. We measured parasitism of R. padi and Aphis glycines by these F 1 females and backcrossed 29 F 1 females with males from the control populations, which would produce backcross females with genotypes that were homozygous control or heterozygous control/selection. We measured parasitism of R. padi and Aphis glycines by exposing them to 387 backcross females and then genotyped 372 backcross females that were recovered live (Table 1).

| Measurement of parasitism
We measured parasitism of R. padi by 100 females of A. rhamni from selection populations and by 20 females from control populations for generations 2 and 3 of selection and parasitism of R. padi and Aphis glycines by 60 and 57 F 1 females, respectively, and 387 and 107 backcross females, respectively. Lastly, we measured parasitism of Aphis glycines by 28 females from control populations. We measured parasitism of Aphis glycines to determine whether improved performance on R. padi correlated with reduced performance on Aphis glycines.
To measure parasitism, we exposed individual female parasitoids to R. padi or Aphis glycines. We used females that were 1-5 days old and had been with males and aphids since emergence and thus had the opportunity to mate, host feed and oviposit. To ensure that females had a full egg load, we isolated females from aphids for 24 hours before using them in experiments. We put each female in a cage (10 cm diameter by 22 cm tall) enclosing the foliage of potted plants with 100 aphids of mixed instars. Female parasitoids were removed either after 24 hours for R. padi or after seven days for Aphis glycines. Ten days later, we collected any mummified aphids and held them for adult parasitoid emergence. After the adults emerged, we recorded the number of mummified aphids and the number of adultparasitoid progeny.
Because A. rhamni females carry about 15 eggs, which they can replace in one day, the abundance of aphids and period of exposure allowed parasitoids to use their full egg complement.
Furthermore, the density of aphids, amount of plant material and cage size meant that parasitoids were unlikely to be limited by search rate. Therefore, we measured a combination of acceptance of hosts for oviposition and suitability of hosts for parasitoid development.

| Analysis of parasitism and adult emergences
Replicates in which females were not recovered or died before the end of the exposure period were not included in analyses because of the risk that they were exposed to aphids for longer or shorter periods than the recovered females. This left 83-96 per cent of females tested (Table 1).
We used generalized linear models (GLMs) to test the effects of selection on the number of parasitized (mummified) aphids of each species. Although we collected data on adult emergence rates (proportion of parasitized aphids from which adult wasps emerged), there were too few R. padi parasitized by control population females to compare rates between treatments. The experimental unit for these analyses was a female parasitoid exposed to a single aphid species. These variables could have non-normal distributions with variances proportional to means, so we used the appropriate error distribution (e.g., normal, negative binomial) for each analysis. We chose the distribution that gave highest model probability calculated from the residual deviance divided by residual degrees of freedom compared with a chi-square distribution (Littell et al., 1996). The negative binomial distribution gave the best fit for the numbers of parasitized aphids. For these analyses, we used the GLM.NB function in the MASS R package (version 7.3-48; Venables & Ripley, 2002) and the glm function in the STATS package in R. We calculated least-squares means and 95% asymptotic confidence intervals using the LSMEANS function in the EMMEANS R package (version 2.27-61; Lenth, 2016). The confidence intervals were sometimes asymmetrical so we report means and asymptotic 95% confidence levels in the following format: mean [lower confidence level − upper confidence level].
Because selection may erode genetic variance, we compared variances for the females from selection generation 2 and 3 using the F-ratio in the var.test function in the R stats package (R_Core_ Team, 2020). We also compared variances among females from selection generation 3, control females and backcross females because heterosis may increase variance among backcross individuals, compared with variances among their progenitors.

| Analysis of associations of phenotypes with SNP loci
Reduced-representation libraries reads were cleaned and trimmed for quality using Trimmamotic (Bolger et al., 2014). The reads were then aligned to a draft genome of A. rhamni using bwa (Li & Durbin, 2009

| Genome assembly and annotation
For de novo assembly of the genome of A. rhamni, we used an Illumina paired-end library (~300 bp inserts with 2 × 150 nt sequencing in one Illumina channel) prepared and sequenced with standard with kits and protocols from Illumina (Illumina). We assembled the genome with MaSuRCA (Zimin et al., 2013) and evaluated the genome assembly with the quantiles of contig sizes, by comparing assembly size to that estimated from flow cytometry, and by comparing gene content with the core insect gene set in BUSCO (Simão et al., 2015).
Using AUGUSTUS with the Nasonia gene model (Stanke & Morgenstern, 2005), we identified protein-coding regions in the A. rhamni genome assembly. To confirm that these genes were transcribed, we mapped RNAseq data to the putative genes using Magic-BLAST (Boratyn et al., 2019). To discover the function of these genes, we compared their amino acid sequences to proteins in the RefSeq database (accessed on 4/21/2018; ncbi.nlm.nih.gov) using BLASTP (with the BLOSUM62 scoring matrix, E-value = 0.001 and the default values for other parameters) (Altschul et al., 1990) and searched for functional information using BLAST2GO (Conesa et al., 2005) and domain analyses with InterProScan (version 5; Jones et al., 2014).

| Functions of genes near loci associated with R. padi parasitism
To explore the relationship between candidate genes and gene functions and genetic markers associated with parasitism of R. padi, we mapped all 184 SNP loci, whether associated with parasitism or not, to the contigs in our A. rhamni assembly. We then separated the contigs into sets that had loci associated with parasitism at FDR ≤0.05, and those that had loci not associated with parasitism at FDR >0.05 or FDR >0.20, and identified the genes on these sets of contigs. This filtering was done with custom scripts in R (R_Core_Team, 2020).
We identified the genes in these sets of contigs, searched for candidate genes (i.e., those coding for venom proteins, cytochrome p450 proteins and chemosensory proteins) and determined whether the sets differed in the numbers of candidate genes. We also tested for enrichment of biological processes or molecular functions between the gene sets using Fisher's exact test (false discovery rate = 0.05) in BLAST2GO.

| Response to selection
In generations 2 and 3, A. rhamni females from the selected populations parasitized 6 and 8 times more R. padi than those from the control populations, but parasitism did not increase between generations 2 and 3 ( Figure 1; Table 2). Surprisingly, 40 to 50 per cent of females from the selected populations failed to produce any offspring on R. padi after two and three generations of rearing on R. padi. This suggests there are recessive alleles for failure to oviposit or survive in R. padi that are slow to be removed by selection. F 1 females from crosses between selection-population males and control-population females parasitized threefold more R. padi than control-population females, but similar numbers to generation 3 selection-population females. Backcross females from the cross of F 1 females with control-population males parasitized threefold more R. padi than control-population females. F 1 and backcross females parasitized more than two-fold fewer Aphis glycines than control females, suggesting trade-offs between parasitism of R. padi and Aphis glycines. Nonetheless, F 1 females parasitized fivefold more Aphis glycines than R. padi, and backcross females parasitized three-fold more Aphis glycines than R. padi. That the difference in parasitism of R. padi versus Aphis glycines is less for backcross females than for the F 1 females is not surprising, given that about half of the backcross females should be homozygous for control alleles, whereas all Assuming an initially linear response to selection, the response in parasitism of R. padi to selection over two generations selection was R 2 = 1.9 aphids (i.e., the difference between the means of generation-2 and control-population means) and the selection differential over two generations was S1 + S2 = 1.95 aphids (i.e., 2 × 0.98, which is the difference between the selected-parent mean and the unselected mean) so the narrow-sense heritability can be estimated as 0.97, which is quite high and explains the rapid response to selection. Among these 180 SNP loci, 28 (16 per cent) were associated with differences in parasitism of R. padi with FDR ≤0.05 (Table 3). Alleles in the selected population at 18 loci increased parasitism over control alleles by 0.8-1.6 aphids, and alleles at 10 loci decreased parasitism of R. padi by 0.9-1.3 aphids (Figure 2

| Genome assembly and annotation
With 25 Gb of paired-end Illumina reads that gave 52× coverage, our assembly of the A. rhamni genome was 384 Mb long and thus 20% smaller than the 483 Mb genome size estimated from flow cytometry (Gokhman et al., 2017). The difference in estimates of genome size between flow cytometry and assembly may result from repetitive DNA, which is difficult to assemble. Our assembly had an N 50 of 18 Kb and 42k contigs with lengths ≥1 Kb. Despite the fragmentation of the assembly, it captured an almost complete set of insect genes, as measured by comparison with the 1658 genes in the BUSCO core insect-gene set (Simão et al., 2015). Our assembly included 98 per cent of the core set, with 96 per cent complete genes, of which 95 per cent were single copy and 1 per cent were duplicated, 2 per cent of the core genes were fragmented, and only 2 per cent of the core genes were missing.
Using AUGUSTUS, we found 41,066 genes, whose combined

| Functions of genes near SNP loci associated with R. padi parasitism
We mapped the 184 SNP loci segregating between control and selection populations to 171 contigs in our draft A. rhamni genome assembly. We separated the contigs into three sets based on the false discovery rates of the SNP loci for association with parasitism of R. padi: FDR ≤0.05, FDR >0.05 or FDR >0.20, and these sets differed in effect sizes (Figure 2). Note that the FDR >0.05 set includes the FDR >0.20 set. The 28 SNP loci associated with parasitism of R. padi mapped to 28 contigs (Table 4), only one of which harboured more than one SNP locus. Together, these contigs comprised a total of 1.2 Mb, and as pointed out above, the loci and thus the contigs mapped to a single linkage group. The three sets of contigs had different numbers and lengths and harboured different numbers of genes (Table 4, Figure 3). We will not consider further the 56 contigs without genes. One long contig (167 kb) had two SNP loci, one of which was associated with parasitism of R. padi (FDR ≤0.05) and the other of which was not (FDR >0.05); however, the latter had p = 0.05 and effect size of 0.7, so we included the 10 genes from this contig in the category of those associated with parasitism of R. padi. Of the 28 contigs having loci associated with parasitism, 9 (median length = 16 kb) had no identified genes, but 19 (median length = 38 kb) harboured a total of 80 genes with most contigs having a single gene, a median of three genes per contig, and a maximum 17 genes per contig (Figure 3). Among these 80 genes, 74 had homologs among insect genes in the nr GenBank database, with all homologs in species of parasitic Hymenoptera, primarily Nasonia vitripennis. Furthermore, 69 genes could be assigned functional annotations with either BLAST2GO or InterProScan.
None of the 74 genes with insect homologs located on contigs with SNP loci associated with parasitism of R. padi were among the candidates expected to affect host specificity, that is those coding for chemosensory, venom or cytochrome p450 proteins. However, among the 373 genes on contigs with SNP loci that had FDR >0.05 for effects on parasitism, there were one odorant-binding protein and three ionotropic receptors, one of which had an effect size of 0.9, p = 0.02 and FDR = 0.09 and so is a marginal candidate.
We compared the biological processes and molecular functions of the set of genes on contigs with SNP loci-associated differences in parasitism of R. padi (FDR ≤0.05) with those of the genes found on contigs with SNP loci definitely not associated with parasitism (FDR >0.20). Among biological processes, there were 18 specific processes and 41 more general processes that either increased or decreased among genes on contigs with SNP loci associated with parasitism compared with those that were not (Fisher's exact test: p ≤ 0.05; Figure 4). Among molecular functions, there were six specific functions and 11 more general functions that either increased or decreased among genes on contigs with SNP loci associated with parasitism (Fisher's exact test: p ≤ 0.05; Figure 5). However, with FDR = 0.05 to correct for multiple testing, we found no enrichment in biological processes or molecular functions between genes on contigs with SNP loci associated with parasitism of R. padi versus genes on contigs with SNP loci not associated with parasitism.

| D ISCUSS I ON
Aphelinus rhamni responded rapidly to selection for parasitism of R. padi, a suboptimal, low-preference host. Selection for R. padi did not lead to parasitism at the level found with Aphis glycines, the original host of A. rhamni, but did lead to lower parasitism of this host by F 1 and backcross females. There was no increase in parasitism of R. padi from generation 2 to generation 3 of selection, suggesting reduced variance available for selection. Indeed, even by generation 3 of selection, about 50 per cent of females failed to produce any progeny on R. padi, which suggests persistent, recessive alleles reducing parasitism of this aphid. The presence of dominance is also supported by expected versus observed means of F 1 and backcross females. The expected value for parasitism of R. padi by F 1 females is the average of the means for control and selection females, that is 1.06 aphids parasitized, but the observed value was 1.28 aphids parasitized, which is 21 per cent higher than the expected value.
Furthermore, the expected value for parasitism of R. padi by backcross females is the average of the mean of control females and 50 per cent of the mean of selection females, that is 1.28 aphids parasitized, but the observed value was 1.47 aphids parasitized, which is 14 per cent higher than the expected value.
The increase in parasitism from rearing on R. padi is genetic rather than the result of conditioning, because F 1 females parasitized five-fold more R. padi than control females even though both sets of females were reared on Aphis glycines.
The genetic architecture of the response to selection appears complex, involving many genes that produce similar phenotypes, based on the effect sizes of the 28 SNP loci spread among 28 genomic contigs. However, all of these contigs mapped to a single linkage group and together comprised only 1.9 Mb, suggesting that a relatively small region of the A. rhamni 483 Mb genome was involved in the response to selection.
We identified 80 genes on the 19 contigs with SNP loci associated with R. padi parasitism, with the majority of these contigs having three or fewer genes. With one exception of an ionotropic receptor on a contig with a marginally significant SNP locus, none of these genes had annotations like those we expected, that is those coding for chemosensory, venom, or cytochrome p450 proteins.  Furthermore, although there were differences in biological processes and molecular functions of these genes and those on contigs with SNP loci not associated with parasitism of R. padi, when a correction of multiple comparison was used, the differences disappeared.
However, there are several caveats concerning our results. First, although we found 28 SNP loci associated with R. padi parasitism on 28 contigs, our mapping population was on the small side, which limits the sampling of recombination events and the power to detect loci with small effects. All 184 SNP loci mapped to a single linkage group comprising 6.5 Mb. Some hitchhiking of regions with no effect on parasitism is likely, given that there had not been many meioses by generation 3.
However, although all the SNP loci were in one linkage group and so on a single chromosome, some were associated with parasitism of R. padi and some were not, so there appears to have been sufficient recombination within the linkage group for loci to have segregated. Inability to detect small effects means we may not have identified all the loci associated with differences in parasitism. Another caveat concerns gene function. We found only 151 chemoreceptor proteins in A. rhamni. By contrast, Nasonia vitripennis, a parasitoid with a well-annotated genome that is smaller than that of A. rhamni based on flow cytometry estimates (Gokhman et al., 2017), has a rich complex of 272 chemoreceptor proteins (Robertson et al., 2010). There may be more chemoreceptor genes in A. rhamni species, but because such genes evolve rapidly, they can be difficult to identify by homology-based searches (Sanchez-Gracia et al., 2009). Thus, additional chemoreceptor genes may exist among those without blast hits or gene ontology annotations. Another and perhaps more valid approach to determining function would be to analyse tissue-or cell-specific expression of candidate genes of unknown function, which we are pursuing.
These results provide one of the few studies of the genetic architecture of host specificity in parasitic wasps. However, research on host specificity in Nasonia species indicates a similar architecture, where a single region of the genome explained differences in host specificity between Nasonia vitripennis, a generalist, and Nasonia giraulti, a specialist. Desjardins et al. (2010) identified a 16-Mb region that when introgressed from N. vitripennis into N. giraulti, altered its host specificity, and more precise mapping has further delineated this to a 4.1-Mb region (Leung, 2020). chromosome organization macromolecule catabolic process cellular macromolecule catabolic process carbohydrate derivative metabolic process negative regulation of macromolecule biosynthetic process negative regulation of cellular biosynthetic process negative regulation of biosynthetic process negative regulation of cellular macromolecule biosynthetic process chromatin organization behavior small molecule catabolic process nervous system process carboxylic acid catabolic process organic acid catabolic process heart development mRNA metabolic process imaginal disc-derived leg morphogenesis circulatory system development renal system process cell fate determination response to insecticide GO Name

| CON CLUS IONS
The implications of our results for the evolutionary shifts of host specificity are somewhat equivocal. Although there was a rapid response to selection for parasitism of R. padi, the levels remained much lower than those for parasitism of its original host, Aphis glycines, with F 1 females of the cross between selection and control populations parasitizing five-fold fewer R. padi than Aphis glycines.
We have continued to rear the selection populations on R. padi, and preliminary results after over 140 generations of selection show that the selection population females still parasitize less than half as many R. padi as Aphis glycines.

CO N FLI C T O F I NTE R E S T
The authors declare no conflict of interest related to this manuscript.