Greenbug aphid [Schizaphis graminum (Rondani)] is a serious insect pest of small grain crops (Royer et al., 2015). It reduces yields by direct feeding and by transmitting plant-pathogenic viruses such as Cereal Yellow Dwarf Virus (CYDV). At the turn of the century, economic losses to greenbug exceeded $100 million per year to wheat and sorghum in the southern Great Plains of the United States (Eddleman et al., 1999). There exist several sources of resistance to greenbug, including the Gb gene in Aegilops (Zhu et al., 2005) and Rsg1 and Rsg2 in barley (Armstrong et al., 2016; Xu et al., 2023). Greenbug variants, traditionally termed biotypes, have been defined by their ability to infest different wheat and sorghum cultivars (Porter et al., 1997; Bowling et al., 1998). Currently 12 biotypes have been designated A through K and TX1 (Armstrong et al., 2016), but a thorough worldwide study of greenbug populations had not been conducted, and possibly many more biotypes remain to be discovered. Greenbug gene expression has been examined by RNA-seq in relation to biotype, time, and CYDV carrier status (Crane et al., 2023).

Aphid microbiomes exist in several sites: the gut, the specialized bacteriocyte cells, the hemolymph, the exterior cuticle, and the immediately local phylloplane. Each has its own conditions and potentially supports a distinct microflora. Almost all aphid taxa harbor Buchnera aphidicola, an intracellular mutualist bacterium that is restricted to bacteriocytes and is vertically transmitted from mother to offspring (Braendle et al., 2003). Other bacteria, such as Hamiltonella defensa (Gupta and Nair, 2020), Rickettsia sp. (Deng et al., 2019), and Candidatus Fukatsuia symbiotica (Michalik et al., 2014; Manzano-Marín et al., 2023) have also been found to inhabit bacteriocytes, sometimes with Buchnera and sometimes separately. The roles of these endosymbionts in aphid nutrition, reproduction, and defense against fungi and parasitoids have been extensively investigated (Csorba et al., 2024). However, the gut is where the most generalist and diverse microflora is to be expected and where a diverse community of apparent commensals exists (Guyomar et al., 2018). Here the microflora is in direct contact with the aphid’s nutritionally deficient diet of plant sap, and here the properties of the host plant can directly affect the competition among microbial taxa (Wilson and Duncan, 2015). These properties include carbohydrate and protein concentration, amounts of antimicrobial peptides and phenolics, pH, and levels of minerals and vitamins (Casteel and Hansen, 2014; Gupta and Nair, 2020; Yun et al., 2014).

Amplification of hypervariable regions within 16S rRNA genes is a popular and effective way to census bacterial populations (Johnson et al., 2019), since the 16S locus is universally present in bacteria (Geng et al., 2022; Haro et al., 2021), highly conserved primer sequences exist (Haro et al., 2021; Thijs et al., 2017), large databases of 16S amplicon sequences are available, the method does not depend upon gene expression, and there are software packages (e.g., QIIME2) specifically designed to process such sequences. However, the method is not perfect, since repeated amplification cycles can magnify the effect of slight variation in primer binding and elongation efficiency, and since some bacterial genomes are present in high copy numbers per cell (notably including Buchnera aphidicola; Komaki and Ishikawa, 1999). Amplicon sequencing will miss any taxa whose 16S hypervariable regions cannot be amplified for any reason.

Previous studies of aphid microbiomes vary in the amount of variation reported within species and among localities. Only a subset deals with host plant as an experimental variable. In general, reports that include rare taxa find more variation among localities or host plants than reports that exclude rare taxa. This reflects the generally high fraction of Buchnera among read counts. When Buchnera comprises 70% or more of reads, there is less opportunity for rare taxa to reach a 1% threshold of abundance that is often used to filter out possible contaminants.

On the basis of Bray–Curtis dissimilarities among 70 samples obtained from 44 host aphid species, McLean et al. (2019) concluded that aphid microbiomes were more correlated with aphid species than with host plant, and this correlation was highly significant for both secondary symbionts and non-symbionts. They found a lesser correlation between aphid microbiomes and host plant habit (tree versus herb) that could reflect the taxonomic relationships of the aphids that infest each. McLean et al. (2019) also noted that the abundance and diversity of non-symbiotic microbes were less in aphids than in insects with richer diets. Microbiome correlation to aphid taxon is necessarily higher in aphid taxa that seasonally alternate between hosts. For example, Wu et al. (2018) reported that 171 microbial operational taxonomic units (OTUs) were shared between summer and winter generations of Schlechtendalia chinensis, while 12 OTUs were restricted to winter and 11 were restricted to summer. The winter and summer host plants differ radically: a moss (Plagiomnium maximoviczii) versus sumac (Rhus chinensis).

Using field-collected specimens, Gallo-Franco et al. (2019) found that weighted and unweighted UniFrac distances of microbiomes of Aphis gossypii and Myzus persicae were more clustered together within aphid species than within host Capsicum species. Also with field-collected specimens, Fakhour et al. (2018) found greater similarity of non-Buchnera microflora within each of five aphid species than within collection localities. Zepeda-Paulo et al. (2018) found variation among field collections in secondary symbiont frequency in Sitobion avenae but essentially no variation among field collections of Rhopalosiphum padi, where Buchnera accounted for at least 98.5% of read counts. Xu et al. (2021) found little effect of collection locale or host plant on alpha diversity or Bray–Curtis dissimilarity in 92 samples of Myzus persicae field collected from 16 host plant families and 30 sites across China. However, Buchnera aphidicola dominated all of the samples, and seven secondary symbionts accounted for most of the remaining counts of taxa that exceeded 1% of the total in any one sample. Each of these seven genera monopolized the non-Buchnera counts in at least one sample. Therefore, the taxonomic composition of the Myzus microbiome could reflect host plant variation or aphid genetics or location history.

Ma et al. (2021) compared Aphis gossypii field-collected on hibiscus and pomegranate with samples of 50 individuals transferred from hibiscus to cotton, muskmelon, and cucumber and then propagated for 10 generations. Because all of these samples were taken from naturally established populations in the field, they were potentially genetically heterogenous. They found that the microbiomes’ taxonomic composition diverged after transfer to the second host plant species and that the Shannon index (within-sample diversity) increased after transfer to muskmelon and cucumber. Thus, in this case a change in diet did affect the microbiome. It is noteworthy that Ma et al. (2021) did not restrict analysis to taxa with at least 1% abundance in at least one sample, whereas Xu et al. (2021) did. If bacterial ploidy increases during the evolution of obligate endosymbiosis as it did in Buchnera (Komaki and Ishikawa, 1999), it is possible that the combined counts of Buchnera and the newly evolved co-obligate endosymbiont would push the maximum count of any other taxon below 1%.

He et al. (2021) conducted a similar experiment with Myzus persicae, which was transferred from Chinese cabbage (the control species) to eggplant, pepper, or tobacco and allowed to feed and reproduce for 14 days. The mean Shannon index increased approximately three-fold on pepper relative to eggplant, tobacco, or Chinese cabbage. Bray–Curtis dissimilarities clustered together for aphids on Chinese cabbage. The taxonomic composition shifted from abundant Buchnera on Chinese cabbage to abundant Pseudomonas on eggplant and tobacco. Pepper had about half as much Pseudomonas as Buchnera. Quantitative PCR showed that the absolute abundance of Buchnera decreased on eggplant and tobacco and remained about the same on pepper as in Chinese cabbage. The ratio of total bacterial to aphid DNA was approximately constant over the four diets.

Two studies considered the effect of diet on the same aphid population. Ikuze et al. (2024) reported family-level frequencies of bacteria from samples of the same Melanaphis sacchari population placed on Melanaphis resistant and susceptible sorghum inbreds. The microbiome was sampled seven and 14 days later. While the report is sparse with details and has inconsistencies with respect to the families reported versus collection time and host plant resistance, it appears that the resistant and susceptible diets did not significantly affect alpha diversity according to the Chao1 index, which increased from day 7 to day 14 on both diets. The pattern of bacterial family frequencies was similar between diets, but the susceptible diet favored Sphingomonadaceae and Pseudomonadaceae at the expense of Burkholderiaceae and Rhizobiaceae. Cassone et al. (2015) looked specifically at Buchnera transcription and titer in Aphis glycines feeding on healthy soybean plants and soybean plants infected with bean pod mottle virus, which aphids do not transmit. They found reduced transcription and titer on the infected plants, which were associated with reduced aphid fecundity. Buchnera transcription recovered when aphids were moved from infected to healthy soybean plants. They noted that the reduced aphid fitness also occurred on soybean infected with two other viruses and attributed the reduction to the dietary quality of infected soybean phloem.

Because of greenbug’s economic importance, multiple resistance genes have been sought and found in landraces and wild relatives of wheat, barley, and sorghum. Among others, these genes include Gb1 through Gb6 in wheat (Porter et al., 1994), Gbx1 and Gbz in wheat wild relative Aegilops triuncialis (Zhu et al., 2005), Rsg1 and Rsg2 in barley (Xu et al., 2023), and SgR1 in sorghum (Zhang et al., 2024). The mechanisms of action of these resistance genes are generally unknown, although SgR1 encodes a leucine-rich repeat receptor-like protein that is expressed in transgenic Arabidopsis thaliana exposed to aphid feeding (Zhang et al., 2024). Their net effect is to inhibit greenbug growth and reproduction, and in doing so they potentially affect the composition and activity of the greenbug microbiome.

Although the type strain of Buchnera aphidicola was chosen from Schizaphis graminum (Munson et al., 1991) and early genomic studies on B. aphidicola were conducted on the S. graminum strain (Lai and Baumann, 1992; Munson et al., 1993; Rauhbakhsh and Baumann, 1995), the greenbug microbiome seems not to have been reported to date. Here we have investigated how host plant species and cultivar affect the bacterial flora of two greenbug biotypes, E and K, over time. The hosts were 17 cultivars from five species, sorghum, wheat, rye, barley and Aegilops triuncialis. Read counts from the PCR-amplified V5-V7 region of 16S rDNA served as a proxy for bacterial populations.

The experiment addressed several hypotheses: (1) that the microbiome responds to host greenbug genetics, here biotype; (2) that the microbiome similarity is related to phylogenetic relationship of the host plants; (3) that the microbiome responds to host-plant resistance genes; and (4) that the microbiome responds to elapsed time. The null hypothesis in each case is that the microbiome does not respond to that factor. The experiment was not designed to elucidate the mechanism by which each factor influenced the microbiome, in that the host plant was not analyzed for any of the plausible mechanisms (nutrition, physical impediments, antibiosis) that could influence the greenbug microbiome, and the greenbugs themselves were not weighed or measured.

Read counts per sample varied from 301,562 to 853,907, with a mean of 603,858 and median of 608,703. Figure 1 is a rarefaction plot generated by repeated subsampling of quality-filtered read counts and identification by alignment to the in-house database of 16S rRNA sequences. The counts were used before normalization to 301,562 counts across samples, so each curve is limited to its total read count. The least diverse samples saturated more than the more diverse samples, but no sample exhausted its own taxonomic variation. At least 1,150 operational taxonomic units (OTUs) were observed in the most diverse samples. Overall there were 1,511 OTUs representing 78 genera.

Only 15 out of 78 identified genera had over 10,000 total counts (Table 2). Buchnera aphidicola dominated the counts, outnumbering second-place Pseudomonas by a factor of 196 and third-place Rhodanobacter by a factor of 840. Buchnera was also by far the most evenly distributed genus among the conditions and replicates, being the only genus with less than 10% of its total counts in any single condition or replicate. At the other extreme, several genera, such as Deinococcus, Sphingobium, and Luteibacter, were almost entirely restricted to a single replicate despite having thousands of counts (Table 2). With the exception of one unidentified species of Serratia, no OTUs mapped to genera of secondary endosymbionts in aphids (Hamiltonella, Regiella, Arsenophonus, Candidatus Fukatsuia, Rickettsia, Rickettsiella, and Wolbachia). To display the relative abundance of other genera, stacked barplots were prepared with Buchnera counts excluded. Figure 2 gives the fractions of the 30 most abundant genera other than Buchnera in relation to time and aphid biotype over all hosts. Pseudomonas peaked in both biotypes on days 2 and 8, while Rhodanobacter was prominent on day 0 in biotype E and day 4 in biotype K. Enterobacter was most common in biotype E on day 4, and the related genus Escherichia was at its maximum in both biotypes on day 4. In biotype K, Chryseobacterium, Luteibacter, and Pantoea peaked on day 4, Massilia peaked on days 4 and 8, while Bacillus peaked on day 2 and Pedobacter peaked on day 8. Chryseobacterium was also high in biotype E on day 4. Unclassified genera, which were identified only to a higher taxonomic level, were most abundant in both biotypes on day 0.

Figure 3 displays the relative abundances by host plant species, summing over timepoints, cultivars, and aphid biotypes. Sorghum differed from the other four species in having less Pseudomonas and more Rhodanobacter. Luteibacter, Chryseobacterium, Pantoea, Massilia, Escherichia, and “unclassified” were also prominent in sorghum. The other four host plant species are all related and belong to the tribe Triticeae, and their non-Buchnera microbiomes were dominated by Pseudomonas. In barley, Enterobacter, Massilia, Pedobacter, Pantoea, Brevundimonas, and Bacillus were consecutively less prominent. In rye, Chryseobacterium, Stenotrophomonas, Massilia, Curtobacterium, and Pantoea were consecutively less abundant. In Aegilops, Propionibacterium, Arthrobacter, Methylophilus, Staphylococcus, and Paenarthrobacter were consecutively less abundant. In wheat, Massilia, Arthrobacter, Burkholderia, Rhodanobacter, and Janthinobacterium were consecutively less abundant.

Relative abundances of non-Buchnera genera are barplotted by cultivar in Figure 4. Pseudomonas dominated in 13 of 17 cultivars. Four cultivars (Wintermalt barley, TX2737 and TX2783 sorghum, and Largo wheat) stood out because of their relative deficiency of Pseudomonas and varying abundance of other genera. In Wintermalt, Enterobacter, Massilia, Pedobacter, and Pantoea were prominent. In TX2737, Rhodanobacter, Luteibacter, and Chryseobacterium were prominent. In TX2783, prominent genera included Massilia, Paracoccus, Burkholderia, and Flavobacterium. In Largo, Paracoccus, Flavobacterium, and Propionibacterium were relatively prominent. There was no correlation of low abundance of Pseudomonas to greenbug biotype resistance among these four cultivars as noted by Armstrong et al. (2016).

Statistical significance of variation in genus counts was appraised with ancombc (Lin and Peddada, 2020). Comparisons of the two biotypes, the four timepoints, and the five host plant species are reported in Table 3, which lists beta coefficients (the slope of the plot of log of taxon abundance divided by days after infestation), their standard error, Holm-adjusted p-values, and a statistic W that is the quotient of the beta value divided by its standard error. A positive beta slope for biotypes indicates up-population in biotype K over biotype E. A positive beta slope for timepoint indicates up-population at a later collection date versus day zero. A positive beta slope for host plant species indicates minimum population in Aegilops. Only three of the genera listed in the biotype section of Table 3 (Rhodanobacter, Luteibacter, and Pedobacter) are prominent in any bar of Figure 2. In the time section of Table 3, only Bacillus is prominent in a bar (biotype K on day 2) in Figure 2. In the species section of Table 3, four of the five identified genera were also prominent in Figure 3.

With respect to greenbug resistance genes, when ancombc (Lin and Peddada, 2020) was run with a frequency feature table of 479 genera that included Buchnera, there were seven genera with unadjusted p-values less than 0.01 as listed in Table 4. None of these was significant even at p < 0.10 after Benjamini–Hochberg correction for multiple testing. Log₂ fold-changes from susceptible to resistant were modest, ranging from −0.32 for Chryseobacterium to 0.52 for Massilia. Positive and negative log₂ fold-changes were split 274 to 205, so there was an excess of genera that became somewhat more abundant in resistant plants at the expense of Buchnera. Excluding Buchnera changed the result almost imperceptibly in Table 4.

Correlations among frequencies of bacterial species appear in Figure 5. Buchnera was negatively correlated with all other taxa except Pantoea dispersa, Propionibacterium sp., Deinococcus sp., and Chryseobacterium takakiae. The negative correlations were mostly significant at p < 0.001 and were strongest for Pseudomonas (the most abundant genus other than Buchnera) and an unspecified genus of Enterobacteriaceae. Mutually positive correlations were conspicuously clustered in a group of 16 taxa in the lower right quadrant of Figure 5 and a second group of five taxa at the upper left corner of Figure 5. There was an isolated strongly positive correlation between species of Enterobacter and Pantoea. While correlations outside of the two groups were weaker, many were still significant at p < 0.001. Only Deinococcus sp. and Chryseobacterium takakiae were not strongly correlated with other taxa overall.

An analysis of variance was performed on the normalized counts (Table 5). All four variables (biotype, infestation timepoint, host plant species, and host cultivar) significantly affected read counts, as did interactions of biotype with the other three variables. Tukey’s Honest Significant Difference test (Miller, 1981) showed that barley differed in observed OTUs from the other four species at adjusted p-value ≤0.01 (Table 6). Figure 6 demonstrates that barley supported a more diverse microflora. Moreover, all three barley cultivars yielded more OTUs than any other sampled cultivar. Specifically, read counts differed at p < 0.05 for nine pairwise comparisons involving biotype K on barley versus biotype K on the other four species and biotype E on any of the five species (Supplementary Table 1). The differences between day 8 and either of day 0 or day 2 were significant at p < 0.05, and day 8 had fewer OTUs (Supplementary Figure 1).

The mean Shannon diversity was 1.13 in biotype E and 1.19 in biotype K (Supplementary Figure 2). Mean Shannon diversity was 1.09 in Aegilops, 1.12 in rye, 1.14 in wheat, 1.16 in sorghum, and 1.30 in barley (Supplementary Figure 3). Three cultivars (Wintermalt barley, Post 90 barley, and TX2737 sorghum) accounted for the greater diversity in barley and sorghum (Supplementary Figure 4). Shannon diversity was also subjected to ANOVA (Supplementary Table 2), and the only significant differences involved the host plant species, the interaction of host plant species with aphid biotype and host cultivar, and the interaction of host plant species with aphid biotype and time. Tukey’s HSD test yielded six significant comparisons of Shannon diversity, again all involving biotype K on barley with biotype K on rye, wheat, or Aegilops, and biotype E on sorghum, wheat, or Aegilops (Supplementary Table 3 and Supplementary Figure 5). Barley, and only barley, differed significantly at p < 0.05 from the other host plant species in supported Shannon diversity (Supplementary Table 4). Shannon diversity did not vary significantly across days (Supplementary Table 5).

In ANOVA of Hill diversity (effective number of species, Supplementary Table 6), only host plant species and the interaction of host plant species with biotype and cultivar had significant effect. Once again, all significant pairwise comparisons with Tukey’s HSD involved biotype K on barley with biotype K on rye, wheat, or Aegilops, or biotype E on any host. Hill diversity was higher for biotype K than biotype E. For comparisons among species, the respective differences in mean significant Hill diversity versus barley were 1.84 for Aegilops, 1.74 for rye, 1.56 for wheat, and 1.36 for sorghum (Supplementary Table 7), and barley supported the highest diversity (4.9 versus 3.5 in sorghum, 3.3 in wheat, 3.1 in rye, and 3.0 in Aegilops). Cultivars Wintermalt, Post 90, and TX2737, yielded the highest Hill diversity (Supplementary Figure 6), and these were the cultivars with the highest fraction of non-Buchnera counts. There were no significant differences among infestation dates.

Beta diversity (diversity among samples rather than within samples) was assessed with Adonis (Anderson, 2001) using weighted or unweighted UniFrac and Jaccard dissimilarities. In addition, pairwise Adonis tests were run for biotype, infestation date, and species. Results of the global Adonis tests appear in Table 7. Nine of eleven variables and interactions were significant with unweighted UniFrac, the only exceptions involving infestation date and host plant species (Table 7). With weighted UniFrac, only the interactions of biotype with host plant species or with host plant species and cultivars were significant. With Jaccard dissimilarity, the only significant variable or interaction was the interaction among biotype, host plant species, and host cultivar, although the interaction of biotype and host plant species was almost significant (p = 0.052) (Table 7). Pairwise Adonis with unweighted UniFrac gave significant differences between biotypes E and K, between barley and any of the other host plant species, and day 2 versus day 4 or 8. The difference between biotypes is illustrated in Figure 7, where the distributions overlap but mostly differ. Pairwise Adonis with weighted UniFrac also gave significant differences between biotypes E and K, along with significant difference between day 4 and the other days, but there were no significant differences among the host plant species. With Jaccard dissimilarity, pairwise Adonis detected only a significant interaction among biotype, host plant species, and cultivar.

The physiological activity of Buchnera versus the rest of the greenbug microbiome was inferred computationally with PICRUSt2 (Douglas et al., 2020), which output pairs of pathway and count related to the population and activity of each identified bacterial or archaeal taxon. Because of the high fraction of Buchnera reads, Buchnera counts exceeded other counts more than 10-fold. Rankings of pathways by count appear in Table 8 and Supplementary Table 8. Table 8 and Supplementary Table 8 have columns for all genera, Buchnera only, and all genera except Buchnera. Because of the abundance of Buchnera, the ranking of the most expressed pathways overall followed the ranking in Buchnera. The rankings indicated that Buchnera had lost 115 out of 421 pathways noted in other genera, including 112 of the 200 lowest-ranking by count and 85 of the 100 lowest-ranking by count. Thus Buchnera had selectively retained the more important pathways. Table 9 and Supplementary Table 9 contain comparative rankings of pathways ordered by activity over all genera except Buchnera. The most important pathway in the lumen bacteria was aerobic respiration utilizing cytochrome C, which releases hydrogen ions into the bacterial environment. This pathway ranked 175th in Buchnera, where acidifying the bacteriocyte cytoplasm is maladaptive. Earlier steps in respiration were not diminished; the tricarboxylic acid cycle ranked third in Buchnera and 14th elsewhere, while the non-oxidative part of the pentose phosphate pathway ranked fifth in Buchnera and 20th elsewhere. At least nine pathways in the synthesis and salvage of long-chain fatty acids were also diminished in Buchnera, including gondoate biosynthesis (60th in Buchnera, third elsewhere), mycolate biosynthesis (162nd in Buchnera, fourth elsewhere), cis-vaccenate biosynthesis (77th in Buchnera, 11th elsewhere), and fatty acid salvage (196th in Buchnera, 22nd elswhere). However, beta oxidation of fatty acids was increased (first in Buchnera, tenth elsewhere). Synthesis of most amino acids ranked similarly in Buchnera and elsewhere. An example was L-valine biosynthesis, ranked 12th in Buchnera and eighth elsewhere.

A conventional study was conducted with the V5–V7 hypervariable region of the 16S rRNA gene in DNA isolated from intact greenbug aphids on 17 cultivars in five host plant species. There were four timepoints post infestation and two aphid biotypes defined by relative virulence on a set of wheat varieties. As in several other studies of 16S rRNA amplicons from aphid microbiomes, Buchnera aphidicola overwhelmingly dominated the read counts, ranging from 78.24 to 99.99% of the total. In Rhopalosiphum padi, B. aphidicola comprised 98.4% of the counts (Zepeda-Paulo et al., 2018) or 96.4% (Fakhour et al., 2018). In Aphis glycines, B. aphidicola comprised 70.3 to 93.5% of counts among geographic locations (Gallo-Franco et al., 2019). In Aphis gossypii, B. aphidicola was 2.9 to 74 times as counted as the second-place taxon (Ma et al., 2021). In Melanaphis sacchari, B. aphidicola comprised 90 to 99% of counts (Holt et al., 2020).

In all of these cases, including the current study, counts of B. aphidicola were inflated by its high copy number of 50 to 200 genomes per cell (Komaki and Ishikawa, 1999, 2000), versus one per cell in typical stationary-phase bacteria. Thus the populations of B. aphidicola and the more abundant gut bacteria like Pseudomonas might have been relatively similar. The count of Buchnera bacteroids per bacteriocyte also responds to environment (at least temperature; Neiers et al., 2021) and increases as the aphid ages. Another possible influence is unequally efficient amplification of various 16S rRNA loci by the degenerate forward primer used here, or unequal amplification based on the specific amplified sequences themselves. A few percent difference in amplification at each PCR cycle can mount up to several fold or more variation in read counts after 30 cycles. To demonstrate this, a simulation was conducted with 1,000 equally abundant taxa, 30 cycles, and amplification rates defined as 1 + (m – 0.5), where m is the mean of N = 20 to 50 random numbers in (0, 1). This resulted in the top 1% of taxa accounting for 11.5% of total amplicon counts with N = 50, 13.9% of total counts with N = 40, 17.8% with N = 30, and 24.8% with N = 20. The mean difference of amplification rates from 1.000 was 3.26% with N = 50, 3.65% with N = 40, 4.21% with N = 30, and 5.16% with N = 20.

The high fraction of Buchnera read counts generally detected in aphid microbiomes implies that the aphid gut lumen hosts relatively few bacteria. The plant sap from phloem is unlikely to harbor many bacteria in healthy plants, so the gut microflora must proliferate during transit through the digestive tract. How much it proliferates depends upon transit duration as well as antimicrobial substances, if any, ingested or secreted by the aphid. The interval between the start of feeding and the start of honeydew excretion is related to transit time, and this interval is measured in minutes for some aphid species. Its duration in greenbug seems not to have been measured in relation to environmental influences such as temperature and relative humidity.

The cause of the uneven distribution of non-Buchnera counts among treatments and replicates in Table 2 is not known, but the relative constancy of Buchnera counts favors a biological cause rather than differential PCR amplification. Possibly interactions among the microbes led to outbreaks of individual taxa in small groups of aphids, much like epidemics of disease. Also, the non-Buchnera sample size was very small in two biotype-timepoints with the highest fraction of Buchnera counts, and this increased the influence of sampling error. Thus it was stated in the Results that Rhodanobacter was abundant in biotype E on day 0 and biotype K on day 4, but it was stated earlier that Rhodanobacter was almost entirely restricted to a single biotype-timepoint (biotype K on day 4). This apparent contradiction resulted from the sum of non-Buchnera reads being 423 for biotype E on day 0 and 251,205 for biotype K on day 4. Non-Buchnera reads summed to 6,215 in biotype K on day 0 and exceeded 46,000 for the remaining combinations of biotype and timepoint.

Other beneficial symbionts (Serratia symbiotica, Hamiltonella defensa, Regiella insecticola, Wolbachia, Arsenophonus sp., Candidatus Ishikawella capsulata; Bansal et al., 2013; Gauthier et al., 2015; Jousselin et al., 2016; Fakhour et al., 2018; Enders et al., 2022) that have been proposed in other aphid species appeared to be absent in greenbug, since no OTUs mapped to them. We detected eight of the 15 other genera noted by Holt et al. (2020) in Melanaphis sacchari, and two (Pseudomonas and Bacillus) were abundant in our non-Buchnera counts. Among other genera that we found, Ma et al. (2021) reported Acinetobacter, Flavobacterium, and Pantoea in Aphis gossypii, and Enders et al. (2022) reported Cutibacterium and Pantoea from two Aphis species. Several other genera (Rhodanobacter, Chryseobacterium, Luteibacter, and Pedobacter) that were relatively common in the current study appear not to have been reported previously in aphid microfloras.

Microbial contamination of labware and PCR reagents can lead to spurious detection of bacterial taxa that are common in the environment. Genera such as Escherichia, Pelomonas, and Methylobacterium, have appeared in reads of PCR product from blank samples (Stinson et al., 2018). Thus, it is not apparent which low-frequency reads came from the reagents and which if any came from low frequencies of the same genera in DNA samples.

It is obvious from the diversity in reported microbes in aphids that aphid genotype and diet affect the microbiome in aphids. He et al. (2021) provide a detailed account of the effect of diet (Yun et al., 2014). They moved a clone of peach aphids (Myzus persicae) from Chinese cabbage to eggplant, tobacco, or pepper. On cabbage, the top-ranking genera were Buchnera (67.7% of counts), Acinetobacter (7.8%), and Allobaculum (5.8%); Pseudomonas was only 0.6%. On eggplant, the top four were Pseudomonas (69.4%), Buchnera (12.5%), Ralstonia (2.1%), and Burkholderia (1.7%). On tobacco, the top genera were Pseudomonas (62.2%), Buchnera (14.6%), Ralstonia (3.0%), and Burkholderia (2.1%). On pepper the top genera shifted to Buchnera (25.5%), Pseudomonas (11.7%), Ralstonia (8.4%), Burkholderia (5.0%), unassigned Enterobacteriaceae (3.2%), Acinetobacter (2.3%), and Stenotrophomonas (1.7%). Thus, eggplant and tobacco greatly favored Pseudomonas and somewhat suppressed Acinetobacter, and there was a rebalancing of total counts between the gut and the bacteriocytes. The authors did not consider the high ploidy of Buchnera in regard to its counts.

With 22 significantly affected genera (Table 3), greenbug biotype was more important than host species, which affected seven genera, or collection time, which affected only two genera. Accordingly, biotype K on barley had the most diverse bacterial communities that were least overwhelmed by the abundance of Buchnera. The eight-day study interval was insufficient time for the aphids to kill the host plant or produce alate morphs, as would be expected at saturation. Resistance genes, which reduced the size of individual greenbugs and reduced their population, had no consistent effect on the microbiome. Either none of the resistance genes affected the microbiome, or they varied in mode of action and thus did not produce a consistent change in microbial abundance. In concept, biotype could affect interactions of microbes with greenbug immunity, the efficiency with which the greenbugs absorbed nutrients from the gut, or the flow of amino acids into and out of the Buchnera bacteroids. However, another plausible mechanism is historical. Greenbug nymphs receive Buchnera from their mother during embryogenesis and acquire lumen bacteria from the environment after birth. However, in an established aphid population, the environment contains honeydew from older aphids, enabling an indirect sort of vertical transmission of gut bacteria from the founders of the population. The role of population history could be tested by moving populations of each biotype back and forth to areas where the other biotype has fed and defecated, and seeing if the microbiomes converge. Molecular genetic markers could be used to control for biotype replacement within the populations.

Microbiome studies in aphids frequently cite the role of Buchnera in providing essential amino acids and remedying the amino acid imbalance in plant sap. However, Buchnera merely redistributes nitrogen among amino acids; it does not increase the total nitrogen supply. Our study encountered low levels of five genera of typically nitrogen-fixing bacteria (Rhizobium, Bradyrhizobium, Azospirillum, Azospira, and Azohydromonas) in the greenbug microbiome without revealing whether they were in the gut or on the phylloplane, which was inadvertently sampled when the greenbugs were collected. This raises the question of whether appreciable nitrogen fixation occurs in the gut. The PICRUSt2 analysis did not reveal nif genes, indicating that the aphid microbiome does not fix nitrogen. Direct measurement of nitrogen fixation or acetylene reduction might be warranted to answer this question. Bacterial nitrogen fixation has been documented in wood-feeding termites of the family Kalotermitidae (Mullins and Su, 2023).

Despite the established role of Buchnera in supplying essential amino acids that are deficient in phloem sap, the PICRUSt2 results did not indicate increased expression of pathways to produce these amino acids relative to the rest of the microbiome. Nevertheless, there is experimental evidence that synthesis of various amino acids is not feedback inhibited or repressed in Buchnera (Wilcox et al., 2003). The most obvious difference that set Buchnera apart was the suppression of electron transport, which releases hydrogen ions to the exterior of the bacterial cell. This suppression is likely an adaptation to life within bacteriocyte cytoplasm, where acidification would be harmful. The diminution of gondoic, mycolic, and cis-vaccenic acids affects the fluidity and thus permeability of Buchnera cell membranes in seemingly contradictory ways. Diminishing mycolic acid expectedly increases fluidity and permeability, but decreasing gondoic and cis-vaccenic acids expectedly decreases fluidity (Marrakchi et al., 2014). Increased permeability would aid export of amino acids from the bacteroids to the bacteriocytes. Conclusions about the role of Buchnera are affected by the great variation among strains in genome size (from 412 to 646 kb) and gene content (354 to 587) (Chong et al., 2019); the strain in Schizaphis graminum is near the upper end of these ranges and expectedly retains more of the functionality of free-living bacteria.