Evolutionary genomics of APSE: a tailed phage that lysogenically converts the bacterium Hamiltonella defensa into a heritable protective symbiont of aphids

Complete genome assemblies for the A2C, AS3, NY26, ZA17, 5AT, MI47, MI12, and 5D strains of H. defensa were previously generated by establishing clonal A. pisum lines containing each strain and then establishing in vitro cultures of each strain from these aphid lines (Table 1) [30, 40, 56, 59], that allowed for single molecule real-time (SMRT) sequencing without amplification or contamination by DNA from the genomes of the aphid or another abundant endosymbiont (Buchnera) following previously described protocols for library preparation and sequencing [56]. Each H. defensa genome was then assembled as detailed by Chevignon et al. [56]. Each APSE genome in the above strains of H. defensa, the APSE1 genome that was sequenced by van der Wilk et al. [8], and the APSE MEAM and APSE MED genomes from the H. defensa MEAM and MED strains were aligned using MAFFT [76] in Geneious (v.10; Biomatters [77]). Protein similarities were assessed by extracting coding sequences (herein genes) followed by translation and alignment in MAFFT. Genes conserved across all haplotypes were named as designated by van der Wilk et al. [8] for APSE1 while genes present in some but not all haplotypes were named as designated in subsequent studies [8, 56]. To assess whether frame shifts observed in p24 from APSE MEAM potentially disables virus production, we used genome sequencing depth as a proxy by accessing the whole-genome shotgun sequence library from the whitefly species in the Bemisia tabaci complex that is the host for H. defensa MEAM (SRR3180082) [38, 78] and converting the data from SRA format to fastq using the SRA-Toolkit v.2.4.2. We then estimated sequence coverage using bowtie2 v.2.2.6 (implementing the -end-to-end option) [79] by aligning and counting reads that mapped to APSE-MEAM or H. defensa with APSE manually removed (NZ_CP016303.1).

We identified candidate orthologs by searching the NCBI non-redundant (nr) database using BLASTP [80] for each gene in the APSE3 AS3 prophage genome. Searches were limited to either entries classified as Caudovirales or γ-Proteobacteria. Results were downloaded as tab separated values and filtered to find the best hit for each gene predicted in the APSE genome. Additional searches were conducted to identify candidate homologs of genes absent in APSE3 AS3 genome, but present in other APSE genomes, including cytolethal distending toxin (cdtB) found in APSE8 ZA17 and shiga toxin (stxB) found in APSE1. Additional focused BLASTP and BLASTN searches of NCBI whole genome sequence (wgs) and RefSeq genomes databases were used to identify proviral elements that shared multiple homologs with APSE. BLAST searches were conducted during January of 2021.

We used a TBLASTX search with APSE3 AS3 genes serving as queries to identify candidate APSE prophage elements in bacterial genomes. Here we searched NCBI RefSeq genomes and wgs databases targeting bacterial genomes classified as Morganelleaceae, Enterobacteriaceae, Pectobacteriaceae, Yersiniaceae, and Erwiniaceae. Once an APSE-like prophage element was discovered, we manually identified the ends of each by comparison with other APSE genomes and then ascertained the extent of shared homologous regions by visualizing alignments using BRIG [81]. We formally compared each phage-like element to APSE using whole genome TBLASTX (V.2.9.0+) and visualizing the results using EasyFig.py (V.2.2.3) [82]. We used CD-search to identify toxin-encoding genes in the newly described APSE genome from Arsenophonus species [83]. In one instance we found a partial APSE-like element within the genome assembly of Arsenophonus nasoniae str. DSM15247. To obtain a more complete assembly of this APSE-like genome we returned to the original A. nasoniae str. DSM15247 read library. We then used aTRAM to isolate and create a draft de novo assembly of the A. nasonia-APSE genome [84‐86]. Multiple APSE genomes were used as baits for aTRAM to collect all possible variations. Contigs obtained from aTRAM with similarity to H. defensa APSE were assessed for similarity to the APSE3 AS3 genome. Next, we generated a reference-guided assembly of the A. nasoniae-APSE genome using Geneious and both the APSE3 AS3 and APSE8 ZA17 genomes from which we extracted a consensus assembly. We then used the recently re-sequenced A. nasoniae str. FIN genome (NCBI assembly GCF_004768525.1) that was generated using a combination of long and short sequence reads to further assess our reference guided assembly and to determine the relative position of the APSE-like elements in the A. nasoniae genome.

Orthologs of p19, p24, p41, and p45 encoded by different APSEs plus other phages and phage elements in bacteria were identified using BLASTP with full-length sequences retained and short or partial BLAST hits being rejected. In addition to the fully sequenced APSE haplotypes that were the focus of this study, this analysis also identified orthologs in other APSE haplotypes that had previously been partially sequenced [55]. Candidate orthologs were downloaded as nucleotide sequences, checked for length, and pseudogenes were identified. We next aligned each set of orthologs including pseudogenes using Geneious V.10 progressive translation guided alignment (translation table 11, PAM250 match/mismatch scoring matrix, gap open penalty of 12, and gap extension penalty of 3). Alignments including pseudogenes were hand corrected to account for frame shifts that disrupted translation alignment processes. We then used RAxML to infer phylogenetic relationships, which uses the General Time Reversible (GTR) model of nucleotide substitutions with the option to model site heterogeneity using Γ and invariant sites [87]. We first used PartitionFinder2 to determine the best model among models available in RAxML and optimal partitions for estimating free parameters [88]. We used AIC_c to select the optimal substitution model when sample size is small (number of sites divided by maximum number of potential model parameters resulted in a low value; observed ranged from 36 to 120). The GTR + Γ models were found to be the best fit and optimal partitions were identified. We then implemented RAxML (HPC v.8.2.8; random seed = 12,345) to find the best tree with appropriate partition and model. Support for phylogenetic relationships was determined as the percent of 1000 bootstrap replicates that agreed with the best tree. Resulting trees were viewed and figures were generated using FigTree v.1.4.3 (http://​tree.​bio.​ed.​ac.​uk/​software/​figtree/​). We then repeated our phylogenetic analysis using RAxML and optimal model selection with pseudogenes removed. Since recombination could also impact our phylogenetic results, we conducted an additional phylogenetic analysis using the NeighborNet method in SplitsTree4 [89, 90] with alignments lacking pseudogenes.

Sequencing of the MI47, MI12 and 5D strains of H. defensa showed that each contained an APSE in the main host chromosome as a single copy provirus. We then used these APSE genomes together with other sequenced haplotypes [8, 38, 51, 56, 59] to compare overall features (Table 1). Total genome sizes ranged from 36,522 bp for APSE1 MI47 to 39,884 bp for APSE2 NY26 (Fig. 1). Predicted genes further ranged from a low of 41 for APSE3 AS3 to a high of 47 for APSE2 NY26, APSE8 ZA17, and APSE8 5D (Fig. 1). BLASTP searches against the NCBI nr database yielded predicted functions for 37 genes while 13 others were classified as unknowns or hypotheticals (Fig. 1, Table 2). Comparing each predicted gene across haplotypes indicated that APSE2 5AT and APSE2 NY26 were nearly identical to one another at the amino acid level (99.8%), whereas overall identities were lower when shared genes were compared to other haplotypes due to several genes including p45, p36 and p37 (Additional file 1: Fig. S1), which had been noted to vary among APSE haplotypes in other studies [67]. We also detected frameshifts in the major capsid protein gene (p24) in APSE MEAM and APSE MED that exist as proviruses in two H. defensa strains present in closely related whitefly species of the Bemisia tabaci complex named MED and MEAM [38]. These frameshifts suggested p24 is pseudogenized in APSE MEAM and APSE MED which combined with previously identified defects in the regulator protein I and p38 (integrase) from APSE MEAM and MED [38] suggest these prophages are inactive. To further investigate this possibility, we used publicly available shotgun sequencing data generated for B. tabaci MEAM to map reads corresponding to H. defensa and APSE. This analysis indicated H. defensa (340×) and APSE (431×) MEAM were sequenced to similar depth, which further supported this APSE haplotype persists as a prophage but likely produces no particles.

Other studies have noted that gene content and order are largely conserved among APSEs [36, 55, 67], but had not assessed genome organization from the perspective of functional units and module composition, which are characteristic of particular phage groups and suggest evolutionary constraints that maintain certain genes together because of their interactive roles in genome replication, lysogeny, virion formation and other essential functions [68‐70, 91‐100]. Examining our data set from these perspectives indicated that APSEs are organized into two functional units with early genes that have functions in integration, lysogeny and replication (module 1 plus p1 in module 2) being on the negative strand and late genes with functions in genome packaging (p2–p5 in module 2), virulence (module 3) and virion assembly (modules 4) being on the positive strand (Fig. 1, Table 1).

Integrases in temperate phages regulate site-specific recombination between the phage (attP) and bacterial (attB) attachment sites [101]. tRNA genes or sequences adjacent to tRNA genes are also common tailed-phage integration sites [102]. APSE phage (attP) and bacterial (attB) attachment sites were previously identified, with the latter occurring in a single copy tRNA-Arg gene [55]. By comparing each APSE-infected strain of H. defensa to the APSE-free A2C strain we showed that this site of integration (attB) was identical among examined strains (Fig. 1). Phage attachment attP core sequences were also almost identical among APSE haplotypes and located in a non-coding region between p37 (domain 4) and p38 (domain 1) (Fig. 1). Integration of each haplotype disrupted the host tRNA-Arg gene, but comparisons to the A2C genome showed that the left (attL) and right (attR) boundaries of the APSE genome complemented the H. defensa tRNA-Arg sequence, which repaired the host gene (Fig. 1).

Our naming scheme differed from Roüil et al. [67] who named APSEs on the basis of gene content in the toxin domain (module 3) and variation in a domain upstream of the DNA polymerase (p45) in module 1. This resulted in APSE8 being classified as a subtype of APSE2. In contrast, whole genome comparisons across all four modules underlies why we continued to distinguish APSE2 from ASPE8 as distinct haplotypes. For the same reasons, we called the phage variant from MI12 H. defensa strain APSE9 as it too was distinct from other named APSEs across modules.

Given the conservation in gene order and content of modules 1, 2 and 4, we selected APSE3 AS3 as a model haplotype and used BLASTP to ask if any genes shared > 60% identity with predicted products from other fully sequenced viruses. Three genes in module 1, two genes in module 2, and seven genes in module 4 were identified that met this criterion, with each best hit being to another phage assigned to the families Podoviridae or Siphoviridae (Table 2). TBLASTX analysis corroborated previously noted similarities in gene order and content between the virion assembly module of APSEs (module 4) and P22-like phages that infect bacteria in the order Enterobacterales [68] including Salmonella virus HK620, Shigella flexneri phage Sf6, Morganella phage NV18, and S. enterica P22 [14, 71, 103‐105] (Fig. 2A, B). Few similarities were detected between APSEs and these P22-like phages outside of their virion assembly modules (Fig. 2A, B), but similarities in gene order and content were identified between APSE module 1 and 2 in podoviruses that infect hosts outside of the Enterobacterales including Xylella phage Xfas53, Burkholderia phage complex members such as BcepC6B, Bordetella phage BPP-1, and Yersinia enterocolitica phage YeP4 [105‐107] (Table 2; Fig. 2B, C). In contrast, no APSE genes in module 3 shared a similar level of sequence similarity with other known viruses.

We also assessed whether high identity homologs existed in any sequenced bacteria outside of H. defensa since this could suggest the presence of APSE-like prophages or prophage elements. We first considered other aphid symbionts, which like H. defensa, reside in the order Enterobacterales. These included several Arsenophonus spp. (Morganellaceae) and a Sodalis sp. (Pectobacteriaceae) that were already known to encode APSE-like genes [67, 72, 75]. We also included Buchnera and Pantoea (Erwiniaceae), Regiella and Fukatsuia that form a clade with H. defensa within the Yersiniaceae [30], and Serratia that is also in the Yersiniaceae. Best hits (25–96% identities) using BLASTP were largely restricted to Arsenophonus spp., but included three genes from Sodalis glossinidius, Ca. Symbiopectobacterium, and Serratia symbiotica (Table 2). Extending this analysis to other Enterobacterales further identified high identity (> 60%) hits to APSE genes in four genera (Xenorhabdus, Morganella, Proteus, and Providencia) from the family Morganellaceae. Only four APSE genes (p19, p23, p27, p28) shared > 60% identity outside of γ-Proteobacteria with best hits to each being to Mycobacterium tuberculosis (Actinomycetales).

Given the preceding results, we interrogated the de novo genome assemblies available for four of the Arsenophonus spp., in which high identity APSE homologs were detected, from the perspective of both gene content and synteny. A single, small contig in Arsenophonus sp. str. ENCA contained a small syntenic region with coding sequences whose translation products shared high identities with products of the p3, p4 and p5 genes in APSE module 2 (Additional file 1: Table S1; Fig. 3A). In Arsenophonus sp. ex. Aleurodicus floccissimus, one contig contained a colinear block consisting of p5, yd repeat and homologs of all genes in conserved order for APSE module 4 (p17–p37), while a single contig was identified in Arsenophonus sp. ex. Bemisia tabaci Asia II 3 that contained p5, cdtB, and most genes (p17–p28) in conserved order for APSE module 4 (Additional file 1: Table S1; Fig. 3B). We recognized that the assemblies for these Arsenophonus spp. could have captured only part of an APSE genome given each derives from short read data and cannot be fully assembled. We therefore asked if reanalysis could generate additional information. We could only access original data for A. nasoniae DSM15247 which consists of short reads generated by Wilkes et al. [74] plus recently generated long read data from A. nasoniae FIN (NCBI SRA SRS441142 and SRX301737). The new assembly we generated using these data, with APSE3 AS3 and APSE8 ZA17 as references, unambiguously identified two syntenic domains. The first consisted of most but not all genes in APSE modules 1 and 2 in conserved order plus f (lysozyme) and p14 in module 3, while the second domain contained p14 plus most genes in module 4 (p17–p33) that were also in near fully conserved order (Fig. 3C). However, more than 1 Mb of A. nasoniae DNA was present between these domains and the prophage element with extensive similarity to APSE modules 1 and 2 was associated with virion assembly genes not found in APSE. Reexamining the plasmid pSOG3 from S. glossinidius str. morsitans showed that six virion assembly genes (module 4) shared > 60% amino acid identities with corresponding APSE3 genes in module 4 (Additional file 1: Table S1; Fig. 3D). However, all other genes on this plasmid were unrelated to APSEs. The S. glossinidius str. morsitans main chromosome contained a second domain encoding genes that shared significant identities with predicted APSE proteins in modules 1–3 but gene order only weakly resembled an APSE due to the presence of several unrelated bacterial genes or viral genes from other phages (Additional file 1: Table S1; Fig. 3D).

Similar analysis of non-symbiont bacteria in the Morganellaceae detected homologs of APSE genes in colinear blocks corresponding to APSE module 1 and 2 plus a partial module 3 containing holin-lysozyme genes in the genomes of Morganella morganii (this region was associated with virion assembly genes not found in APSE) and two Providencia species (Additional file 1: Table S1; Fig. 4A). Colinear blocks corresponding to APSE module 4 were also identified in M. morganii, Proteus mirabilis, and two other Providencia species (Additional file 1: Table S1; Fig. 4B). In contrast, no colinear blocks or recognizable homologs were identified in these species that corresponded to APSE module 3 outside of p14 and p16. Close inspection of the three intact phages (MmP1, MP1, and MP2) that have been identified from M. morganii [108, 109], four intact phages (PM16, PM75, PM87, and PM93) that have been identified from Proteus mirabilis [110] and a single phage (PR1) identified from Providencia rettgeri [111] indicated that none shared genes or sequence homology with APSEs. A BLASTN search failed to find any additional phages that have been deposited into NCBI which contain APSE-like modules like those present in M. morganii.

Altogether, no fully intact APSE-like genomes were identified outside of H. defensa, but colinear blocks containing high identity genes in syntenic order that corresponded to APSE modules 1, 2 and 4 were in both Arsenophonus spp. that are insect symbionts and certain other species in the Morganellaceae that were not. However, the only bacterium outside of H. defensa that contained a largely intact APSE-like toxin domain (module 3) was Arsenophonus sp. ex. Aleurodicus floccissimus.

Since phage elements with similar gene order in APSE modules 1 and 4 were identified in some symbiont and enteric species of Enterobacterales, we generated phylogenies using two module 1 genes, p41 (helicase) and p45 (DNApol), and two module 4 genes, p19 (Portal protein) and p24 (Major capsid protein). These genes were selected to capture phylogenetic signal from each module and represented genes for which we could easily obtain orthologs from other phage and bacterial genomes, hence providing phylogenetic signal. APSE MEAM and APSE MED exhibit structural mutations in p45 that suggest they are pseudogenized [38] while BLASTP identified frame shift mutations in p19 from Arsenophonus sp. ex. Aleurodicus floccissimus. Phylogenetic trees further suggested inclusion of p45 and p19 pseudogenes generated phylogenetic error; namely long-branch attraction due to increased rates of nucleotide substitution in pseudogenes. We therefore conducted a phylogenetic analysis with pseudogenes removed. This analysis yielded several well-supported relationships (bootstrap support greater than 75%). Using genes from module 1 indicated all APSE haplotypes from H. defensa form a clade that is sister to APSE-like elements in Arsenophonus spp., while genes from module 4 found a similar pattern with homologs from APSE and Arsenophonus spp. being sister to prophage elements in Morganella, Proteus, and Providencia spp. (Fig. 5). Given the possibility for recombination events generating false phylogenetic signals [112, 113], we constructed phylogenetic networks using the same genes, which also supported that APSEs from H. defensa were closest to Arsenophonus spp. (Additional file 1: Fig. S2).

As previously noted, gene content in module 3 differs among haplotypes with APSE1, 4 and 5 encoding stxB toxin subunit genes, APSE2 5AT, APSE2 NY26, APSE6, APSE7, APSE8 ZA17, APSE8 5D APSE MEAM, and APSE MED encoding cdtB toxin subunit genes, and APSE3 AS3 encoding a yd repeat toxin gene [38, 51, 55]. We asked if cdtB represents a plesiomorphy within APSE phages. If cdtB diverged with APSE strains, we would expect the sequences in APSE2 5AT, APSE2 NY26 and APSE8 ZA17, APSE8 5D to share a similar number of identical DNA bases as APSE MEAM and APSE MED when compared to cdtB in, for example, Arsenophonus spp. ex Bemisa tabaci (in this case cdtB is a true homolog). However, if a cdtB moved by horizontal transfer from APSEs that infect H. defensa to the APSE-like phage element in Arsenophonus spp. ex Bemisa tabaci (or vice versa), then we would expect one of the cdtB genes in APSEs from H. defensa to be more similar to the Arsenophonus based cdtB gene (a paralog). We would further predict that cdtB in APSE MEAM/MED would also be more similar to the APSE-like cdtB gene in Arsenophonus spp. ex Bemisa tabaci given each infects the same whitefly host as H. defensa strains MEAM and MED. Results indicated that the percentage of shared identical nucleotides between cdtB in the phage element in Arsenophonus spp. ex. Bemisia tabaci Asia II 3 and APSEs were similar for both APSE MEAM/MED (46.1%) and APSE2/8 (47.7%), which was consistent with the cdtB genes representing a plesiomorphy in APSE (fig. S3).