Impact of Facultative Bacteria on the Metabolic Function of an Obligate Insect-Bacterial Symbiosis

Although microbial colonization of the internal tissues of animals generally causes septicemia and death, various animals are persistently associated with benign or beneficial microorganisms in their blood or internal organs. The metabolic consequences of these persistent associations for the animal host are largely unknown. Our research on the facultative bacterium Hamiltonella, localized primarily to the hemolymph of pea aphids, demonstrated that although Hamiltonella imposed no major reconfiguration of the aphid metabolome, it did alter the metabolic relations between the aphid and its obligate intracellular symbiont, Buchnera. Specifically, Buchnera produced more histidine in Hamiltonella-positive aphids to support both Hamiltonella demand for histidine and Hamiltonella-induced increase in host demand. This study demonstrates how microorganisms associated with internal tissues of animals can influence specific aspects of metabolic interactions between the animal host and co-occurring microorganisms.

). Initial analysis with individual t tests, as is standard for analysis of metabolomics data (43), identified 11 metabolites that differed significantly between Hamiltonellabearing and Hamiltonella-free aphids (see Table S1 in the supplemental material). However, when these 11 metabolites were taken for more rigorous analysis that included the effect of aphid genotype and correction for multiple testing, no metabolites differed significantly between Hamiltonella-bearing and Hamiltonella-free aphids ( Table S1).
Construction of isogenic lines bearing and lacking Hamiltonella. Two factors may contribute to the lack of significant metabolic differences between Hamiltonellabearing and Hamiltonella-free aphid genotypes: Hamiltonella may have a minimal effect on the metabolite pools of the aphids; or the influence of Hamiltonella on aphid metabolite profiles may be obscured by metabolic variation among the aphid-Buchnera genotypes. To control for genotype effects, we selectively eliminated Hamiltonella from aphids of genotype SC_583, yielding SC_583 Hlacking Hamiltonella. (SC_583 was selected because it is amenable to Hamiltonella clearance using antibiotics, and it is one of the genotypes used in the initial metabolomics study.) Neither aphid performance nor the abundance and activity of their Buchnera populations differed significantly between the two lines (Table 1; see also Text S1A and B in the supplemental material). These data indicate that any metabolic differences between the two lines are likely direct effects of Hamiltonella on the metabolic function of the symbiosis, rather than a nonspecific consequence of Hamiltonella effects on aphid growth, development, or reproductive output or on the Buchnera population.  Metabolic profile of isogenic aphids bearing and lacking Hamiltonella. To investigate the metabolic traits of the aphid lines SC_583 and SC_583 H-, we applied LC-MS to measure the metabolite profiles of the 7-day-old aphid larvae that had been reared on chemically defined diets from day 2 (Table S2). Three metabolites, all with roles in phenylalanine metabolism differed significantly between the two lines after correction for multiple tests: hydroxyphenylpyruvate, prephenate, and phenyllactic acid (Table S3). These metabolites had not been identified as candidates in our initial analysis of naturally Hamiltonella-bearing and Hamiltonella-free clones (Table S1).
Further inspection of the metabolomics data yielded just one metabolite, 5-aminoimidazole-4-carboxamide ribonucleotide (AICAR), which had a near-significant enrichment in Hamiltonella-bearing aphids (i.e., significant prior to adjustment for multiple tests) in both data sets (Table S1 and Table S3). AICAR is of considerable interest because its production is linked to the overproduction of the EAA histidine by Buchnera. Specifically, AICAR is a by-product of Buchnera histidine biosynthesis and, due to deletion of the proximal reactions for de novo purine biosynthesis, the sole Buchneraderived substrate for Buchnera purine synthesis ( Fig. 2A). It has been argued that Buchnera demand for AICAR to meet its purine requirements drives the overproduction of histidine, with the excess histidine delivered to the aphid host (44,45). We hypothesized that, in Hamiltonella-bearing aphids, the metabolic demand for Buchneraderived histidine exceeds Buchnera demand for purines, leading to the accumulation of AICAR.
Buchnera-mediated histidine biosynthesis. To investigate the effect of Hamiltonella on Buchnera-mediated histidine biosynthesis, we raised larvae of the lines SC_583 and SC_583 Hon chemically defined diet with histidine supplied exclusively as [ 13 C 6 ]histidine for 5 days and then measured the 13 C/ 12 C ratio of free histidine and protein-bound histidine (Table S4). [  Consistent with our prediction that histidine synthesis is increased in Hamiltonellabearing aphids, the 13 C/ 12 C ratio in the soluble histidine pool was significantly lower in SC_583 than SC_583 Haphids at P ϭ 0.025 threshold (Bonferroni correction for two tests) (Fig. 2B). The equivalent data for histidine in the protein fraction showed the same trend of reduced 13 C/ 12 C in SC_583, but the effect was not significant (Fig. 2B). 13 C was predominantly recovered from fully labeled [ 13 C]histidine (His Mϩ6), accounting for 53 to 63% of total histidine in the soluble fraction and 47 to 55% in the protein fraction, and the equivalent values for fully unlabeled [ 12 C]histidine (His Mϩ0) were 32 to 42% and 39 to 48%, respectively (see Fig. S1 in the supplemental material).
Further analysis of this data set showed that AICAR content, but not histidine content, was significantly elevated in SC_583 relative to SC_583 H- (Fig. 2C), recapitulating the results of the previous experiments (Tables S1 and S3).
The metabolic determinants of AICAR content. We hypothesized that the relationship between increased Buchnera production of histidine (as revealed by the 13 C/ 12 C ratio of histidine) and increased AICAR content of aphids bearing Hamiltonella could be explained by the metabolic link between the synthesis of histidine and purines in Buchnera ( Fig. 2A). Specifically, AICAR is predicted to accumulate under conditions where the total symbiosis demand for Buchnera-derived histidine exceeds the Buchnera demand for AICAR as the substrate for purines.
To investigate whether Hamiltonella demand for extra histidine creates an overflow of AICAR from Buchnera, we compared the metabolic flux in a two-compartment metabolic model, comprising Buchnera and the aphid host, and three-compartment models that also included Hamiltonella with Buchnera/Hamiltonella biomass ratios ranging from 10:1 to 1:5 (Fig. 3). We applied flux balance analysis to quantify how AICAR production varies with histidine synthesis, as determined by flux through the HisD reaction. In the Hamiltonella-free model, Buchnera releases no AICAR (Fig. 3A). In the three-compartment model with Buchnera/Hamiltonella biomass ratio greater than one, Facultative Symbiotic Bacteria and Metabolism ® Buchnera exhibits modest increase in histidine production ( Fig. 3B), supporting Hamiltonella demand for this EAA, accompanied by AICAR overflow (Fig. 3A). As the Hamiltonella biomass exceeds that of Buchnera, the model predicts further increase in Buchnera histidine production and AICAR overflow. At the most extreme Buchnera/ Hamiltonella biomass ratio tested of 1:5 (equivalent to 1:130 cell number ratio), HisD flux is increased by 35% and AICAR overflow is more than doubled. The relative cell number of Buchnera/Hamiltonella in line SC_583 was determined empirically, at 1:3.9 (standard error [s.e.] ϭ 0.3, n ϭ 18), equivalent to the biomass ratio of 6.7:1 (Text S1B). Under these conditions, the predicted increase in histidine biosynthesis to support the Hamiltonella population is 0.8%, and the predicted AICAR overflow is 0.03 mmol g Buchnera biomass Ϫ1 h Ϫ1 (Fig. 3).
The modest increase in histidine yield (0.8%) predicted by our models compared to the 25% increase in histidine pools observed empirically suggests that an increase in Hamiltonella demand for histidine alone may not fully account for the increased Buchnera histidine production observed in Hamiltonella-bearing line SC_583. Altogether, our empirical and modeling data indicate that Hamiltonella induces additional demands for Buchnera histidine production by a third player, the aphid host.

DISCUSSION
Animals that naturally house a few microbial taxa are powerful systems to investigate the metabolic interactions among microbial taxa and the host. Here, we leveraged the tripartite symbiosis between the pea aphid, its obligate nutritional symbiont Buchnera, and a facultative defensive symbiont Hamiltonella to investigate how Hamiltonella affects host-Buchnera metabolic function. We demonstrated that, in the presence of Hamiltonella, Buchnera increases production of the EAA histidine, and we inferred that the extra Buchnera-derived histidine meets both Hamiltonella demand and Hamiltonella-induced increase in host demand for this EAA. Here, in the Discussion, we explore, in turn, how our three approaches, metabolomics, metabolic experiments, and metabolic modeling, contribute to our understanding of the metabolic consequences of Hamiltonella and some implications for metabolic interactions between other symbiotic microorganisms and their animal hosts.
Our first metabolomics analysis revealed a far greater effect of aphid genotype than Hamiltonella on the metabolome of aphids naturally bearing and lacking Hamiltonella (Fig. 1). These results show that our methodology was appropriate to detect substantial metabolomics differences and suggest that Hamiltonella may not cause a major reconfiguration of the host metabolome. Additionally, the clustering of the metabolomics data by genotype is indicative of large-scale intraspecific variation in metabolic function of the pea aphid. This striking pattern complements published evidence for significant among-genotype variation in aphid utilization of sucrose and amino acids (29,(47)(48)(49), the chief carbon and nitrogen sources in the aphid diet of plant phloem sap. Furthermore, this variation has been causally linked to specific genes of the aphid and Buchnera, as well as variation in Buchnera population size (47,(50)(51)(52). These patterns are also fully consistent with evidence from other animals that host genotype can strongly influence metabolic traits linked to microbiome function (53)(54)(55)(56)(57).
Although Hamiltonella does not perturb the global metabolic homeostasis of its aphid host, our metabolomics data sets included one metabolite, AICAR, with elevated titer in Hamiltonella-bearing aphids for both the comparisons between genotypes that naturally harbor and lack Hamiltonella and between a single genotype bearing and experimentally deprived of Hamiltonella. The correspondence between the presence of Hamiltonella, increased AICAR, and increased Buchnera-mediated histidine production, as determined by our 13 C metabolic experiments, confirmed our hypothesis that the additional AICAR was likely of Buchnera origin and linked to increased demand for Buchnera-derived histidine in Hamiltonella-bearing aphids.
Aphids bearing Hamiltonella are expected to have an elevated histidine requirement because Hamiltonella is auxotrophic for this EAA (46,58). Hamiltonella has the genetic capacity to synthesize just two EAAs, threonine and lysine, and it is expected to be a sink for the other eight EAAs (including histidine), all of which are synthesized by Buchnera (24). These predicted EAA fluxes from Buchnera to Hamiltonella do not translate into significant effects on the titers of histidine or other EAAs in the aphid metabolome, presumably because of homeostatic controls over metabolite pool sizes. Hamiltonella may impose a greater demand for histidine than other EAAs; alternatively, increased flux through other Buchnera EAA biosynthetic pathways in Hamiltonellabearing aphids may have gone undetected in our study because it did not result in the accumulation of unique by-product(s) equivalent to AICAR for histidine synthesis. An indication that Hamiltonella may alter the metabolism of a second EAA, phenylalanine, comes from the significant underrepresentation of three intermediates in the phenylalanine biosynthetic and catabolic pathways (hydroxyphenylpyruvate, prephenate and phenyllactic acid) in Hamiltonella-bearing isogenic aphids (see Table S3 in the supplemental material).
Interestingly, the nutritional requirements of Hamiltonella resulted in no discernible reduction of aphid fitness, despite increased EAA demand. Possible contributory factors were that we used naturally occurring aphid-Hamiltonella combinations and a susceptible plant cultivar for insect culture. In published studies, the effect of Hamiltonella on aphid performance varies with aphid and bacterial genotype (59-61) and can be particularly deleterious for aphids reared on partially resistant plants (36,62).
An important inference from this study is that Hamiltonella demand for histidine is unlikely to account fully for the difference in histidine production between Hamiltonella-bearing and Hamiltonella-free lines. The chief evidence came from metabolic models that assumed fixed aphid demand for histidine; when Hamiltonella at the empirically determined biomass was added to the aphid-Buchnera model, the computed flux of histidine synthesis increased by just 0.8%, substantially less than the observed 25% difference in Buchnera-derived histidine production between aphids bearing and lacking Hamiltonella. A possible factor contributing to this discrepancy may have been the simplifying assumptions required to construct the flux balance models (63,64), although the model equations are not discernibly biased to underestimate Hamiltonella demand for histidine. It is most probable that increased host demand for histidine in Hamiltonella-bearing aphids contributes much of the discrepancy between the empirical data ( Fig. 2B) and model data (Fig. 3).
Why might Hamiltonella increase host demand for histidine and possibly other Buchnera-derived EAAs? Two processes may be involved. First, Hamiltonella has been demonstrated to alter the cellular immunity of pea aphids, specifically by increasing the population of hemocytes (37). This effect would increase the host sink for histidine because as for animals generally (65,66), immune cell proliferation in aphids is metabolically demanding and requires metabolic resources, including EAAs. Second, aphid feeding, including probing behavior and food ingestion can be altered by Hamiltonella (36). These feeding traits can substantially affect dietary EAA supply to the aphid by influencing aphid choice of feeding site and food consumption rates. Previous research has demonstrated that rearing aphids on diets lacking a specific EAA results in increased synthesis of that EAA by Buchnera (67). This suggests that increased production of histidine by Buchnera could arise from feeding changes in aphids bearing Hamiltonella that resulted in reduced uptake of dietary histidine.
In conclusion, this study has revealed that the tripartite relationship between two bacterial symbionts and their aphid host is metabolically interactive. Specifically, the nutritional requirements of one bacterium, Hamiltonella, can modify the metabolic function of a second symbiont, Buchnera, for increased histidine production without altering the Buchnera population size, and the Buchnera response to support Hamiltonella is likely compounded by the Hamiltonella-induced increase in host demand for histidine. These findings raise two general questions. The first relates to the role of Hamiltonella as a defensive symbiont that protects the aphid host against parasitic wasps (33). Although this defensive function has been attributed to toxins encoded by a prophage on the Hamiltonella genome (35), future research should consider the possible contribution to parasitoid resistance of Hamiltonella-induced changes to aphid Facultative Symbiotic Bacteria and Metabolism ® metabolism and immunity. The second question is the incidence of multiway metabolic interactions in symbioses. On the one hand, the substantial among-genotype differences in the aphid metabolome identified in this study suggests that further insights can be gleaned from analysis of intraspecific variation in these interactions. This avenue would be a productive extension of recent research on variation in aphid-Buchnera interactions (50,52) and idiosyncratic effects of facultative symbionts on host phenotype (59)(60)(61). On the other hand, the broad principle of multiway interactions may be general to multiple taxa localized to the hemolymph and cells of insects (68), including other facultative symbionts of aphids and bacteria with a broad distribution in arthropods, e.g., Wolbachia, Spiroplasma. As in this study, these future investigations will be facilitated by the combined application of metabolomics, isotope tracer experiments, and metabolic modeling.

MATERIALS AND METHODS
Experimental aphids. The experiments were conducted on six genotypes of the pea aphid Acyrthosiphon pisum collected from an alfalfa field in Ithaca, NY, USA, in May 2015. All genotypes bore the vertically transmitted bacterial symbiont Buchnera; genotypes SC_12, SC_240 and SC_594 bore no previously described facultative symbionts, also known as secondary symbionts, SC_533 and SC_583 bore Hamiltonella, and SC_495 had both Hamiltonella and Spiroplasma (50).
The aphids were maintained on Vicia faba cv. Windsor at 20°C with 16-h light/8-h dark light cycle. To generate age-synchronized larvae, adult apterous females were allowed to larviposit for 24 h on V. faba plants or excised leaves and then removed. The deposited larvae were left to develop for a further day (to day 2) for use in experiments.

Metabolomics (LC-MS) analysis.
To analyze aphid genotypes that naturally harbored and lacked Hamiltonella, five or six replicate pools of mixed-age aphids of each genotype were collected from routine culture on plants, with 25 mg fresh weight per replicate. Following sample preparation (see Text S1C in the supplemental material), 5 l of each sample was injected onto an AB SCIEX 5600 TripleTOF (triple time of flight) liquid chromatography-mass spectrometry (LC-MS) for analysis in positive and negative electrospray ionization (ESI) mode, with method blanks and a pooled quality control (QC) sample. Samples were separated by reverse-phase high-performance liquid chromatography (HPLC) using Prominence 20 UFLCXR system (Shimadzu) with a BEH C 18 column (100 mm ϫ 2.1 mm; 1.7 m particle size; Waters) maintained at 55°C and a 20-min aqueous acetonitrile gradient (flow rate 250 l min Ϫ1 ). The initial conditions were 97% solvent A (HPLC grade water with 0.1% formic acid) and 3% solvent B (HPLC grade acetonitrile with 0.1% formic acid), increasing to 45% solvent B at 10 min, 75% solvent B at 12 min where it was held until 17.5 min before returning to initial conditions. The eluate was delivered into a 5600 (QTOF) TripleTOF using a Duospray ion source (AB SCIEX). The capillary voltage was set at 5.5 kV in positive ion mode and 4.5 kV in negative ion mode, with declustering potential of 80 V. The mass spectrometer was operated in information-dependent acquisition mode with 100-ms survey scan from 100 to 1200 m/z, and up to 20 MS/MS (tandem MS) product ion scans (100 ms) per duty cycle using a collision energy of 50 V with a 20-V spread.
Instrument raw data files were converted into mzML format using Proteowizard (69) and analyzed using MS-DIAL (70). Mass spectrometry tolerances for MS1 and MS2 were set to 0.01 Da and 0.05 Da, corresponding to the resolution of the TripleTOF instrument. For smoothing extracted-ion chromatography, the linear weighted moving average was applied with a smoothing level of 3, and the minimum peak height was set to 3,000 for noise signal. Compound identification used MS/MS similarity to the curated public library in MS-DIAL with 80% similarity threshold and peak areas normalized using the internal standard chlorpropamide (Santa Cruz Biotech).
Metabolomics analysis of isogenic lines SC_583 and SC_583 Hused three groups of 10 2-day-old larvae administered diet with 2 mM histidine to mimic V. faba phloem sap (71). Five days later, the larvae (7-days-old, approximately 25 mg fresh aphid material per sample) were snap-frozen in liquid nitrogen and stored at -80°C. Following extraction of metabolites (protocol in Text S1D), metabolites were separated using a Waters XSelect HSS T3 column (100 Å, 5 m, 2.1 mm ϫ 100 mm) fitted with a Restex UltraShield 0.2-m precolumn filter and analyzed on a Thermo Exactive Plus Orbitrap (72)(73)(74). Blanks and a pooled QC sample were included in the analysis. Raw data files were converted to mzXML format and processed in MAVEN (75) using an in-house targeted metabolite library (72). Compounds were identified based on m/z (Ϯ10 ppm) and retention time (Ϯ0.5 min) tolerances to standards. Peak areas were normalized against the total ion chromatogram to account for analytical drift, followed by blank subtraction.
For analysis of dietary [ 13 C]histidine metabolism by isogenic aphid lines SC_583 and SC_583 H-, six replicate groups of 2-day-old aphids per line were raised on diets containing 2 mM 13 C 6 -labeled histidine for 5 days (to day 7). Each replicate, comprising 30 larvae (approximately 25 mg fresh aphid material), was snap-frozen in liquid nitrogen and stored at -80°C. Three replicates per line were analyzed for soluble metabolites and extracted as described above, and protein hydrolysates were prepared as described previously (76) (Text S1E). Samples were analyzed on a Thermo Exactive Plus Orbitrap as described above. For histidine isotopes, all isotopic peaks were picked manually using the expected isotopic mass and the tolerances m/z (Ϯ10 ppm) and retention time (Ϯ0.5 min).

Blow et al.
® Antibiotic treatment of pea aphid genotype SC_583 to eliminate Hamiltonella. The larvae of genotype SC_583 were treated with antibiotics as in reference 29. Three cages, each with five 2-day-old larvae of genotype SC_583, were maintained on a chemically defined diet (formulation in reference 77 with 0.5 M sucrose and 0.15 M amino acids), supplemented with the antibiotics gentamicin, cefotaxime, and ampicillin (Sigma) each at 50 g ml ؊1 diet for 5 days (to day 7), when they were transferred individually to plants and allowed to larviposit.
A single offspring (generation 2) per aphid was tested for Hamiltonella by diagnostic PCR assay (50) (Text S1F). Up to 10 siblings of individuals that tested negative were transferred individually to fresh plants, and five of their progeny (generation 3) were tested by PCR for Hamiltonella. This was repeated to generation 10. Where an aphid tested positive for Hamiltonella, all codescendants from the generation 2 aphid were discarded. By this procedure, we generated the Hamiltonella-free line SC_583 H-. As expected, hemolymph samples from leg bleeds of genotype SC_583 bore many bacterial cells of the morphology predicted for Hamiltonella (21, 31,) but hemolymph samples from line SC_583 Hwere bacteria free.
Aphid performance assays. To determine larval relative growth rate (RGR) of aphid lines SC_583 and SC_583 H-, 15 replicate groups of five 2-day-old aphids were weighed (Ϯ1 g) and confined to clip cages on V. faba plants. On day 7, the surviving larvae in each clip cage were counted and weighed. RGR was calculated as log e (day 7 weight per aphid/day 2 weight per aphid)/5 days. To determine the intrinsic rate of increase (r m ), larvae deposited by adults over 24 h on V. faba plants were allowed to develop for 6 days, when they were in the final larval stadium, and then individually transferred to the underside of a leaf of a fresh plant in a clip cage. Aphids were then examined daily until the first offspring was deposited, to give the time from larviposition to onset of reproduction (d). Progeny were subsequently counted and removed every several days until 2d, yielding the total number of progeny per aphid (Md). The formula r m ϭ 0.745(log e M d )/d (78) was applied, with 10 replicate individuals per line.

Metabolic model reconstruction and analysis.
A genome scale metabolic model of Hamiltonella defensa was generated by combining two draft model reconstructions. The first identified gene orthologs in the Hamiltonella defensa genome (NCBI JAABOV000000000) and Escherichia coli strain K-12 substrain MG1655 by reciprocal BLAST searches and then manually extracted reactions encoded by these genes from E. coli strain K-12 substrain MG1655 metabolic model iML1515 (79) to create a draft model, as previously described (41). The second draft reconstruction was generated from the automated reconstruction pipeline ModelSEED (80) using a RAST (81) reannotated Hamiltonella defensa genome as input. The two draft models were then integrated and manually curated to remove redundant reactions and ensure correct reaction gene association, directionality, stoichiometry, and mass-charge balance. Hamiltonella-specific features and genes encoding metabolic reactions absent in the E. coli iML1515 metabolic model were identified by literature review and searches of the BioCyc, KEGG, EcoCyc, BiGG, and BRENDA databases (82)(83)(84)(85)(86) and then added to the draft model. The Buchnera metabolic model was updated from the published model (44,87) by removing two reactions, adding 62 reactions (Data Set S3A) and updating the biomass equation (see Data Set S3B for details).
A genome scale metabolic model for the aphid host was generated as previously described (41), using aphid reactions involved in primary metabolism identified from publicly available Acyrthosiphon pisum genome data (NCBI: GCA_000142985.2). Additional reactions to generate or consume dead-end metabolites were identified and incorporated into the aphid host draft reconstruction. Individual bacterial and host metabolic models were integrated into a three-compartment model using previously described methods (41,88). Model testing was conducted in COBRA Toolbox version 3.0 (89) run in Matlab 2015b (The MathWorks Inc.), using the Gurobi version 6.5.0 solver (Gurobi Optimization 2016). Details of model analysis and constraints applied are provided in Text S1G.
Statistical analyses. All statistical analyses were performed in R (90). Data sets were inspected for normality and homoscedasticity using Shapiro-Wilk and Kolomogorov-Smirnov tests and Bartlett's test, respectively. Differences in data with normal distributions and homogenous variances were tested with Student's t test, and data sets with nonnormal distributions and/or heterogeneous variances were tested with Mann-Whitney U test. Statistically significant differences in metabolite abundance detected by LC-MS analysis of six aphid genotypes were calculated using a linear mixed-effects model (LMM) using the "lmer" function in the lme4 package, version 1.1-19 (91). Hamiltonella status was treated as a fixed effect, and aphid genotype was treated as a random effect. Residuals were visually inspected for normality, and an analysis of variance (ANOVA) was performed using the car package, version 3.0-2 (92). P values were adjusted for multiple testing using the Benjamini-Hochberg method.
Data availability. The Hamiltonella defensa genome assembly and raw sequencing reads used to generate the genome scale metabolic model are available in the GenBank repository under accession number PRJNA602159 (https://www.ncbi.nlm.nih.gov/bioproject/PRJNA602159). The multicompartment model is provided in three formats-SBML (.xml), MATLAB (.mat), and Excel (.xls)-and deposited in GitHub (https://github.com/na423/Aphid_symbiosis). An SBML file of the models is also available in the BioModels database (93) with the identifier MODEL2001310002. All other data are provided in supplemental files.

SUPPLEMENTAL MATERIAL
Supplemental material is available online only. TEXT S1, DOCX file, 0.02 MB. FIG S1, PDF

ACKNOWLEDGMENTS
We thank Mihee Choi for assistance with aphid maintenance and Mariyam Masood for help with aphid sample preparation for DNA sequencing.