Expression Dynamics of the Aplysia Abyssovirus

Abstract

Two recent studies documented the genome of a novel, extremely large (35.9 kb), nidovirus in RNA sequence databases from the marine neural model Aplysia californica. The goal of the present study was to document the distribution and transcriptional dynamics of this virus, Aplysia abyssovirus 1 (AAbV), in maricultured and wild animals. We confirmed previous findings that AAbV RNA is widespread and reaches extraordinary levels in apparently healthy animals. Transmission electron microscopy identified viral replication factories in ciliated gill epithelial cells but not in neurons where viral RNA is most highly expressed. Viral transcripts do not exhibit evidence of discontinuous RNA synthesis as in coronaviruses but are consistent with production of a single leaderless subgenomic RNA, as in the Gill-associated virus of Penaeus monodon. Splicing patterns in chronically infected adults suggested high levels of defective genomes, possibly explaining the lack of obvious disease signs in high viral load animals.

Introduction

The marine gastropod Aplysia californica is a widely used neurological model. The nervous system of A. californica comprise roughly 20,000, giant, identifiable neurons arranged in simple neural circuits (Strumwasser, 1971). The large size and highly polyploid nature of A. californica neurons allows for neurophysiological and molecular analyses at the cellular level (Moroz, 2011b).

In 2018, two data mining studies identified and described a potentially novel virus in publicly available A. californica RNA sequencing databases (Bukhari et al., 2018; Debat, 2018). Spanning more than 35 kilobases, this novel virus, dubbed Aplysia abyssovirus 1 (AAbV) by Bukhari et al. 2018, pushes the limits of genome size in RNA viruses. Initial investigation of tissue distribution of the virus from public datasets suggested AAbV has broad viral tropism with high enrichment in the nervous system. Based upon sequence homology, AAbV was assigned as the only known member of a new family, the Abyssoviridae, in a new suborder, the Abnidovirineae, in the Nidovirales order of positive stranded RNA viruses.

The order Nidovirales includes several clades of viruses including Arteriviridae, Roniviridae, and Coronaviridae (Enjuanes et al., 2008; Gorbalenya et al., 2006). While the most well-known nidoviruses are respiratory or gastrointestinal viruses of terrestrial vertebrates (Burrell et al., 2017; Snijder et al., 2013; Stenglein et al., 2014; Tokarz et al., 2015), the family also contains viruses of fish and invertebrates (Cowley, 2016). Several nidoviruses exhibit neurotropism as has been suggested for AAbV (Dunowska et al., 2012; Lavi et al., 1999). The murine hepatitis virus (MHV), a coronavirus species, is a well-studied model for neurotropic and persistent neural infection (Adami et al., 1995; Haring and Perlman, 2001; Lavi et al., 1984; Lavi et al., 1987; Rowe et al., 1997; Wang et al., 1990). Given the role of A. californica as a premier neural model and the known capacity for neurotropism in nidoviruses, understanding the biology of AAbV is imperative for continued neurobiological research in A. californica.

While the sequence, putative genes, function of the main protease and translational stop-suppression signals (Bukhari et al., 2018) and phylogeny of AAbV has been described, there have yet to be any studies that document the growth of this virus in vivo. Little is known about the transcriptional dynamics, etiology, and possible pathogenesis of this virus. Furthermore, the original description of AAbV relied on data mining of previously published transcriptional databases whose data derive from mariculture specimens reared at the National Resource for Aplysia (NRA) at the University of Miami. Thus, it is unknown if this virus is present in wild specimens.

To address these questions, we mined all A. californica transcriptomic databases publicly available from the NCBI short read archive, including two large datasets not available at the initial description of AAbV and a newly generated RNAseq dataset to identify the patterns of viral tropism, expression levels, and transcriptional dynamics of AAbV. Our analysis was informed by what is known about transcription of well-studied nidoviruses, particularly coronaviruses such as SARS-CoV-2, namely the use of transcription regulatory sequence sites, sub-genomic RNAs (sgRNA), and canonical splicing via leader and body sequences (Nomburg et al., 2020). We further supplemented these data with qPCR analyses of hatchery bred and wild caught animals.

Materials and Methods

Animals

Captive bred A. californica of known age and parentage were obtained from the National Resource for Aplysia. Animals were designated based on the sibling cohort (ie, from a single egg mass from a single animal). Wild animals were obtained from Marinus Scientific LLC, collected in the area of Long Beach, California. Animals were anaesthetized via injection of 1/6 bodyweight of 365 mM MgCl2. Once the anesthetic had taken effect (approx. 5 minutes), animals were sacrificed by complete removal of the nervous system. Animals were sampled for the following tissues: abdominal ganglion (A), all head ganglia pooled (G), anterior tentacle (T), gill (L), heart (H), I-4 muscle (I), ovotestis (O), hepatopancreas (P), and a section of gut just behind the crop (G). Tissues were immediately stored in 0.5ml of TriZol reagent. TriZol was allowed to perfuse the tissues for several hours and then samples were frozen at −20C for long term storage.

Transmission Electron Microscopy

Dissected tissue samples for electron microscopy were fixed in 2% glutaraldehyde in filtered seawater buffered with 0.05 M sodium cacodylate (Electron Microscopy Sciences) and stored at 4°C. Fixed samples were then immersed in three changes of the buffer, post-fixed in 1% OsO₄, dehydrated through a series of graded ethanol concentrations (20%–100%), and embedded in a mixture of EM-bed/Araldite (Electron Microscopy Sciences). Thick sections (~1 μm) were then stained with Richardson’s stain for observation by light microscopy. Ultrathin (~100 nm) sections were cut on a Leica Ultracut-R ultramicrotome and stained with uranyl acetate and lead citrate. Grids were viewed at 80 kV in a JEOL JEM-1400 transmission electron microscope, or at 100 kV in a JEOL 1200 EX transmission electron microscope with images captured on a bottom-mounted 3k x 3k slowscan, lens-coupled CCD camera (SIA 15C).

RNA Extraction and cDNA Synthesis

RNA was extracted from tissues using TriZol phenol-chloroform method according to manufacturer protocol. Samples were DNased using TURBO DNA-free (Invitrogen) and RNA levels quantified by Nanodrop. cDNA was synthesized from 200 ng RNA using the GoScript Reverse Transcriptase kit (Promega) with random hexamers (NEB) to yield 20 ul.

Quantitative Real Time PCR

Viral RNA levels were quantified using the SYBRGreen qPCR kit. 1 ul of cDNA (equivalent to 10 ng starting RNA) was used as input for each reaction. Quantification was performed on a Quantstudio 3 Real Time PCR system (ThermoFisher Scientific). Viral copy number was calculated using a standard curve for each primer and normalized per ng of input RNA per reaction. Viral load was assed using three primers each targeting a different region of the viral genome: (Table 1).

Table 1. Primers for qPCR.

AV4 primer targets the far 5` end of the genome in the ORF1 region, Pol1 primer targeting the viral polymerase gene near the center of the viral genome in the ORF1B region, and A9 targeting the far 3` region of the viral genome in the ORF2 region.

Primer	Sequence (5’->3’)	Start	Stop
AV4-F	GTAAAAGCGCAGAGCGGATG	543	562
AV4 -R	CCATTCAGCGTCCTGTGGAT	684	665
Pol1-F	CGACCACCTTCAGACCAAGT	17791	17810
Pol1-R	GTTTACCTCCCTACCTACACTCTG	17915	17938
A9-F	TGTTTCTGGTCATCAAATTGCTGT	32625	32648
A9-R	AAAGACAGCATCCTATTCTTTTGAG	32738	32714

RNA Sequencing

Total RNA samples were sent to GeneWiz/Azenta Life Sciences for library preparation and sequencing. RNA was sequenced in three phases. RNA originating from abdominal ganglia were prepared using a PolyA-selection unstranded NEBNext Ultra II RNA Library Prep kit (30–488489843). RNA originating from gill was prepared using a PolyA-selection NEBNext Ultra II Directional RNA Library Prep Kit (30–626240528). RNA from several samples used in the previous two phases were resubmitted to for sequencing after ribodepletion with a Ribo-zero Plus kit (Illumina). All samples were sequenced on an Illumina HiSeq2500.

Structural Domain Prediction

The genome of AAbV was analyzed to better understand the domain repertoire of the structural polyprotein. TMHMM 2.0 (Krogh et al. 2001) was used to predict transmembrane regions in the structural polyprotein. Protein homology and structure were predicted using HHpred (Soding et al. 2005). The remaining domains were provisionally named according to similarity of JPred (Drozdetskiy et al. 2015) predicted secondary structure with other known RNA virus structural proteins. Polyprotein sequences were searched for sequence similarity to other viruses against the NCBI sequence archive using BLAST (Camacho et al. 2009).

Raw Read Processing and Alignment

In addition to newly sequenced data, raw reads from publicly available RNA sequencing studies on Aplysia californica were downloaded from the Short Read Archive (SRA) database using the sra-tools software utility (https://github.com/ncbi/sra-tools). Adapters and contaminating sequences were trimmed from raw reads using BBDuk from the BBTools software suite (Bushnell 2014). Reads were then quality filtered and trimmed using sickle (Joshi and Fass 2011) and re-paired using repair from BBtools. Properly paired reads were aligned to an updated AAbV genome using the STAR aligner (Dobin et al. 2013) with parameters specified for viral genomes, as described by (Nomburg et al., 2020). Resulting BAM files were then used for downstream analyses.

Viral Genome Coverage

Per base coverage along the viral genome was calculated from each alignment file using the depth command from samtools (Li et al. 2009). Coverage of specific regions, including viral genes and qPCR primer target regions, was calculated using the multicov function from BEDTools (Quinlan and Hall 2010) and a custom BED file (supplemental data BED). Strandedness of libraries was determined using the infer_experiment.py script from the RNASeQC software suite (Graubert et al. 2021).

Splice Sites and Leader Sequence

The software package virORF_direct (https://github.com/jnoms/virORF_direct), which was designed to identify TRS sites in SARS-CoV2 (Nomburg et al., 2020), was used to identify any potential evidence of a leader sequence in AAbV reads. The generate_synthetic_transcripts.py, find_trs_sites.py, and find_trs_lengths.py, were modified for AAbV features as opposed to SARS-CoV2 features. De novo predicted TRS sites were compared to manually identified putative TRS sites (table 2). Splice sites identified by virORF_direct from STAR alignments to the AAbV genomes were used for splice site analysis.

Table 2. Putative TRS sites.

Manually predicted leader and body transcription regulatory sequences (TRS-L, TRS-B) from AAbV genome. Bolded and underlined sections of sequence represent regions of perfect complementarity. Underlined but not bolded nucleotides represent possible G-T wobble pairs in reverse complement. With the wobble pair, the leader and body sequences share an 11-nucleotide sequence of exact match.

TRS	Start	Stop	Sequence
Leader	42	84	CACAAACCCAAACAAAAAACTTAAACTGCCAAATTTTGGCTT
Body	25154	25196	CCAAAATATCTTAGTTCTAAAAATTTAAAGTGGCTCCATGCG

Data Processing and Visualization

Data were processed and visualized in the R statistical environment with Rstudio using the tidyverse (Wickham et al 2019), GridExtra (Augie and Antonov 2017), and Zoo packages (Zeileis and Grothendieck 2005). A full list of packages and versions used can be found in supplemental file SF1.

Results

Viral distribution and tropism

Evidence of AAbV was detected in 209 of the 222 publicly available Aplysia RNA sequencing samples surveyed (15 and 114 of these samples were used by Bukhari et al. 2018 and Debat 2018 respectively). Fastq files for four of the 222 samples available were improperly formatted and could not be repaired and so were removed from further analysis (SRR027169, SRR027053, SRR027057, SRR027059). Samples were drawn from 12 studies available on the NCBI SRA. These sequencing studies represent diverse tissues and levels of tissue specificity, ranging from whole animals down to single cells and cellular compartments (Supplemental File SF2). Of 209 samples, 9 samples had 0 reads mapped to virus. These samples were a single sample drawn from PVC sensory neurons, two samples drawn from single neurons (L7 and R2), a single sample drawn from sensory neuron process, and 5 samples drawn from Fragile X or Kinesin cargos. Notably, only the studies sampling Fragile X cargo (SRP001195) and Kinesin cargo (SRP001196) did not contain samples detectable viral reads (data not shown).

Viral reads could even be detected in cytoplasmic and nuclear fractions of neurons from study SRP180130 (and even at low levels in individual neuronal processes in study SRP001188) (Figure 1). Re-assembly of viral reads also revealed strong support for an additional 41 nt at the 5’-terminus of the genome that had been omitted from previous reconstructions, 5’-CGGGACTTTATAAAAAATTTCTTAGTATCTTATATATAGTA-3’. All numbering in this manuscript is based on the extended genome sequence

Figure 1. Viral Reads from Cellular Fractions of individual Aplysia neurons.

Raw reads were filtered and mapped to the AAbV genome with STAR. Viral Counts per million (vCPM) were calculated by dividing the sum of uniquely mapped reads and multi-mapped reads by total reads input into STAR and multiplied by 1,000,000. Stacked bars represent individual runs and are colored according to the cellular fraction from which samples were taken as reported in the metadata file for SRP180130 and SRP001188. Data from SRP180130 represent RNA sequences from individual compartments of the same single giant neuron, either the abdominal R2 or left pleural LPL1 motor neurons. Data from SRP001188 represents RNA sequences from individual processes of abdominal LE sensory neurons. Viral loads were higher on average in the cytoplasmic fraction than the nuclear fraction.

When analyzing the 2011 Broad Institute whole transcriptome dataset (PRJNA77701), which sampled 10 different tissues, we identified viral reads in all sampled tissues (Figure 2). The viral load, calculated as viral reads per million input reads (vCPM) was dramatically higher in nervous system samples than all other tissues, with heart samples representing a distant second. Samples from gill, muscle, and anterior tentacles/rhinophores had similar, intermediate viral load. Viral loads were roughly an order of magnitude lower in samples from the digestive system, hepatopancreas, and hermaphroditic gland (Figure 2).

Figure 2. Tissue Level Viral Tropism.

Raw reads were filtered and mapped to the AAbV genome with STAR. Counts per million (vCPM) were calculated by dividing the sum of uniquely mapped reads and multi-mapped reads by total reads input into STAR and multiplied by 1,000,000. Bars are colorized according to the tissue from which samples were taken as reported in the metadata file for SRP009481. The nervous system exhibited dramatically higher viral load than all other tissues, with heart being second highest.

Viral tropism in different neuron types in the nervous system could be assessed using samples from study SRP007385 which sampled either individual neurons or small numbers of individual neurons in each sample (Figure 3). Viral reads were detected in all 11 neuron types from that study, with viral loads in the tens of thousands CPM in 10 of the 27 samples; 22 CPM was the lowest viral load. In fact, of the 161 samples from 6 studies that sampled the nervous system, viral load was detected in 154 different samples originating from 4 different ganglia (pleural, buccal, abdominal, and cerebral) and representing 13 different neuron types.

Figure 3. Viral Tropism of Neuron Types.

Raw reads were filtered and mapped to the AAbV genome with STAR. Counts per million (vCPM) were calculated by dividing the sum of uniquely mapped reads and multi-mapped reads by total reads input into STAR and multiplied by 1,000,000. Bars are colorized according to the neural cell clusters or individual neurons from which samples were taken as reported in the metadata file for SRP000986 and SRP007385. The number of cells pooled in one run is represented by the shape at the top of each bar as reported in the run metadata. Viral tropism is diverse in the nervous system, with viral reads detectable in 11 different neuron types.

To validate the organ level tropism observed in the Broad Institute data, viral RNA was quantified across 8 tissues in animals from several cohorts reared at the National Resource for Aplysia using qPCR (Figure 4). The qPCR data demonstrated that indeed viral tropism is generally highest in nervous system, followed by the heart and gill, although high levels were sporadically detected in ovotestis as well. Gut, I-4 muscle, and anterior tentacles tended to exhibit lower viral loads. These trends are broadly similar to those of the Broad Institute RNA sequencing data.

Figure 4. Tissue Level Viral Tropism via qPCR.

Tissue samples from 4 maricultured F1 cohorts (B10, B15, B23, and B31) and wild animals were evaluated for AAbV viral load using qPCR. Tissue types color coded as indicated in the legend. Nervous system samples (abdominal and head ganglia) generally exhibited highest viral loads, often with heart representing a close third. Gill was also often high in viral RNAs. Generally, gut, I-4 muscle, and ovotestis were lower in viral RNAs. Wild animals exhibited highly variable levels of viral RNA with roughly similar tropism in the higher viral load animals to the patterns seen in cultured animals.

To test whether this pathogen originated in the mariculture system at the National Resource for Aplysia, wild animals from California that had not yet entered the mariculture system were also assessed for viral RNAs with qPCR (Figure 4). Wild animals exhibited a wide range of viral loads from relatively low levels (less than 10 copies per ng RNA) to very high levels, similar to those seen in some of the hatchery reared animals (> 100,000 copies per ng RNA). Among the highest viral load animals, patterns of tissue distribution of virus were similar to those in hatchery animals with abdominal and head ganglia generally the highest, while gill, heart and ovotestis were highly variable. These results suggest that this virus is a naturally occurring in the native habitat of Aplysia californica and that it is a common component of their natural life history. Together these data demonstrate that this virus is both naturally occurring and that viral load is consistently high in the nervous system and highly variable in other tissues of these animals.

Transmission Electron Microscopy

Electron microscopy identified structures resembling double membrane vesicle (DMV) replication factories that appear to contain AAbV virions in ciliated gill epithelium cells. Structures resembling pleomorphic particles known from coronavirus infection were also identified (Figure 5). Extensive imaging of ganglia did not reveal recognizable virus-induced structures. We also note that the pleomorphic particles of variable internal content noted as alphavirus-like by Prince et al., (Prince, 2003) and identified speculatively as AAbV virions by Bukhari et al (Bukhari et al., 2018) were present in roughly equal proportion in ciliated epithelial cells from the gill of animals with high (31/56; 55%), medium (21/28; 75%), and low (20/41; 49%) viral loads in abdominal ganglion, four head ganglia, and gill tissue, suggesting that those particles are likely not associated with AAbV infection. However, the lack of a completely AAbV negative individual prevented us from being able to definitively conclude that the pleomorphic intracellular particles are not virus-associated. Furthermore, these structures cannot be unambiguously attributed to AAbV with EM alone; more robust methods such as immune-EM are required for conclusive positive identification. Nevertheless, these micrographs suggest active infection at the very least in the respiratory organ of Aplysia. Despite much higher virus loads detected by qPCR in ganglia, virus-induced intracellular features similar to those seen in other nidovirus infections were not detected, though the implications of that finding are not clear.

Figure 5. Comparison of intracellular features associated with AAbV and model nidovirus infection.

Electron micrographs show Aplysia abyssovirus 1 (A, C, D) infected ciliated epithelium in the gill of two chronically infected adult Aplysia californica, with Human coronavirus 229E (B, E) in human fetal lung fibroblast MRC5 cells 12h after inoculation for comparison. Characteristic virus-associated features of double-membrane vesicles (blue) and virions in intracellular compartments (red) are marked with colored overlays (A-B). Magnified images (C-E) show fibrous material inside the double-membrane vesicles.

Structural Domain Prediction

Analysis with TMHMM revealed three potential transmembrane regions near the amino terminus of the structural polyprotein, followed by five pairs of closely-spaced transmembrane regions. This organization resembled that of the alphavirus structural polyprotein, with protein ectodomains preceding a double-transmembrane region, and proteolytic processing within or after the paired transmembrane helices. The paired transmembrane regions were treated as provisional domain-dividers, in lieu of other bioinformatic landmarks. HHpred analysis detected homology to a trypsin-like protease in the domain between the third and fourth predicted transmembrane region (for example, e = 2.2 × 10–22 match to 4F4O, the structure of a porcine serine protease involved in haemoglobin scavenging), but no other homology was detected. The carboxyl-terminal region after the last predicted pair of transmembrane regions displayed a high arginine content, indicative of a basic charge, as observed in nucleoproteins of coronaviruses, for example, and was termed the N-like domain. None of the domains matched significantly to any database tested in BLAST. These domain names and boundaries are only intended as placeholders, until functional assays can be done, and are not intended to convey structural or functional information.

Effects of Sequencing Technology

The publicly available Aplysia RNA sequencing data used in this study spans several sequencing technologies. When examining the proportion of input reads mapping to virus, in the overwhelming majority fewer than 5% of total reads originated from virus (supplemental file SF2). Only 3 samples had between 5% and 10% of input reads mapping to virus. Surprisingly, 6 samples had greater than 10% of input reads mapping to virus, with one sample having 41.5% of input reads mapping to virus. All samples with greater than 10% reads mapping to virus were 12-month-old animals from the study SRP136115. This study represents the only study to use Ribosomal depletion (ribodepletion) from among the publicly available datasets. This made it impossible to determine if library preparation methodology was causal for the extreme enrichment of viral RNAs in these samples. Thus, we sequenced several samples using both PolyA enrichment and Ribodepletion for direct comparison. Comparing the proportion of reads originating from virus in the same RNA sample, sequencing strategy effects detection of viral transcripts from total RNA, with some samples reaching greater than 20% of input reads mapping to virus when using Ribodepletion (Figure 6). In fact, Ribodepleted samples exhibited on median 20 times the viral reads of PolyA enriched samples for both gill and abdominal ganglia. To determine if the massive size of the viral genome (~35kb) contributed to proportionally higher levels of viral RNA measured in ribodepleted samples, the proportion of host reads mapping to an extremely long endogenous mRNA species, apTwitchin (XM_013086125.2; 17,394 bp long), was also calculated. The proportion of reads mapping to apTwitchin between PolyA and ribodepleted samples was essentially identical for the gill samples and highly variable in the samples from the abdominal ganglia, with neither technology showing a consistent pattern of enrichment (Figure 6). This suggests that some other factor beyond RNA size contributed to the much greater capture efficiency of viral RNAs by ribodepletion.

Figure 6. Percent of Input Reads originating from virus based on RNAseq library selection method.

Libraries generated from the same RNA samples (abdominal ganglion or gill) using either ribodepletion or poly-A selection resulted in different proportions of total input reads aligning to AAbV (A). On median, the proportion of total input reads aligning to the AAbV genome was 20 times greater in ribodepleted libraries than poly-A enriched samples. This is in contrast to the difference in proportion of host reads mapping to apTwitchin (XM_013086125.2; 17,394 bp long) in Ribodepleted and PolyA selected samples (B). This suggests that the length of the viral genome is not the driving factor of viral read enrichment in Ribodepleted samples.

General trends in high vs low viral loads based on all three qPCR primers and ribodepleted RNA libraries tracked relatively closely. PolyA libraries showed less correspondence (Figure 7). Rank is more consistent in gill samples than in abdominal ganglion samples, which may be due to the lower total viral load in gill samples. However, the variability in rank further demonstrates the limited capacity of comparison among these methods beyond broad relative categories such as “high” and “low” viral load. Between the two technologies, qPCR and RNAseq, the A9 primer and Ribodepleted samples show most similarity and rank in abdominal ganglion.

Figure 7. Viral load rank by quantification method.

RNAseq data (polyA, RiboDep) shown as viral reads per million input reads, qPCR data (AV4, Pol1, A9) calculated as viral copies per ng of input RNA. Samples are colored and linked by colored line across methods for ease of comparison. High and low viral load samples are generally of similar rank between methods, however samples 20B27A and 2B27A show high variability in rank.

Viral Genome Coverage

Whereas qPCR primers provided a limited snapshot of the data, the RNAseq data provided full genomic coverage. Comparison of per base coverage along the viral genome between PolyA and Ribodepleted data revealed strong 3’ depth bias in PolyA samples compared to the rest of the viral genome (Figure 8). Ribodepleted samples exhibited enrichment in both the 5` and 3` ends of similar magnitude. Compared to Ribodepletion and qPCR results, PolyA data was proportionally more enriched for the ORF2 region of the genome which encodes sgRNAs in other nidoviruses (Figure 9). Furthermore, per base depth calculations did not identify a several orders of magnitude level of enrichment of the ORF2 region relative to ORF1 in AAbV, based on Poly A or ribodepleted RNA sequencing, as is typical for infection in coronaviruses. The median per base depth of the ORF2 region was roughly 2 times higher than that of the ORF1 region, with the regions corresponding to the carboxyl-terminal putative N-like region of the structural polyprotein exhibiting higher coverage than the rest of the ORF2 region, potentially as a result of inclusion of these regions in defective genomes (see below).

Figure 8. Viral genome coverage using PolyA vs Ribodepletion.

Per base coverage for the three samples with highest viral load that were sequenced with both Ribodepleted libraries and PolyA selected libraries (2B27A, 17B27A, 19B27A, vCPM > 30,000). Sequencing depth at each position was calculated from alignment BAMs using samtools. The median expression value at each position was taken among the three samples for PolyA selected (black trace) and RiboDepleted (Blue Trace) sequencing runs. A schematic of the major regions of the AAbV genome (ORF1a, ORF1b, ORF2, and 3` UTR) is shown above the traces. Predicted subgenomic RNAs based on sequence homology to other nidoviruses are demarcated as colored regions [Trypsin-like serine protease (T, green), S-like protein (S, cyan), Alphavirus_E1-like_protein (E1, yellow), Alphavirus_E2-like_protein (E2, gold), M-like protein (M, olive), and nucleocapsid-like protein (N, magenta)]. Putative leader and body transcription regulation sites (TRS-B, TRS-L) are marked with a “^”. Per base coverage exhibits a roughly 2-fold increase in the ORF2 region, with slightly increased coverage also evident in the first 5 kilobases in both PolyA selected and RiboDepleted runs. Red vertical line marks genomic position 25133 which contains a region of complementarity between TRS-B and TRS-L corresponding to the beginning of coverage increase in ORF2.

Figure 9. Ratio of target regions of AAbV genome among sequencing technologies.

Viral load was calculated for regions targeted by qPCR primers from samples which were quantified using qPCR, Poly A enriched RNAseq, and Ribodepleted RNAseq. Samples derive from both Abdominal ganglion and Gill tissue. Only samples quantified with all three methods were considered for comparison. Hits to each region were normalized to the hits for the Pol1 primer. Both qPCR and Ribodepleted samples show a bias for the 5’ region of the viral genome (AV4 primer), while Poly A samples exhibit a bias for the 3’ region of the viral genome (A9 primer). These data demonstrated quantification technology introduces a technical bias when quantifying viral load.

Strand Polarity

Strand specificity of publicly available (SRP136115 and SRP268155) and newly generated datasets (Gill polyA and all Ribodepleted libraires) allowed for calculation of strand of origin for viral reads. Viral reads originated overwhelmingly ( > 90%) from the positive strand in the majority of samples. In PolyA selected samples with low viral load (e.g. <10 CPM), the proportion of reads originating from the negative strand was higher, up to 100% in one sample (Figure 10). Strand polarity of total host reads as well as GAPDH specifically remained stable in across samples, while stand polarity of viral reads depended on viral load (Figure 11).

Figure 10. Viral read strand of origin.

Proportion of reads originating from the positive strand were calculated from samples from publicly available (SRP68155 and SRP136115) and newly generated (Schamle_Poly_A_Gill and Schmale_RiboDep) datasets generated with strand specific libraries. Plotting the proportion of reads from positive strand against per sample viral load reveals that samples with high viral load (e.g. >100 CPM) contain predominantly (>90%) positive sense reads. In some low viral load samples, proportion of reads originating from negative strand may be as high as 100%.

Figure 11. Strand of origin of total host, viral, and reference gene reads.

Samples from SRP268155, which were generated with strand specific and poly-A selected libraries, were assessed for the proportion of reads originating from the negative strand for all host reads (teal), GAPDH aligning reads (purple), and all viral reads (yellow). Total host aligning reads originated from the negative strand on average 50% of the time, while GAPDH reads originated from the positive strand nearly 100% of the time as expected. More than 90% of viral reads originated from the positive strand except in some samples with low viral load where reads originated from the negative strand 50–100% of the time.

An important note is that for experiment SRP136115, read files SRR6871213, SRR6871214, SRR6871216, SRR6871221, SRR6871223, SRR6871224 were uploaded such that mate 1 was labeled as mate 2 and vice versa. This resulted in half the files being identified as first strand positive by RNASeQC instead of first strand negative as would be expected for the Illumina kit used. This was corrected locally.

Splice Sites and Leader Sequence

Splice junction analysis of viral transcripts showed no evidence of any canonical splice sites in AAbV (Figure 12). Similarly, TRS site analysis did not identify any incorporation of putative leader or body sequences into spliced reads. Putative TRS sequences generated de novo from splicing data with virORF_direct did not identify any seed sequence common to all studies, tissues, or samples; further suggesting the absence of a canonical leader-body splicing event typical of coronaviruses, and suggesting a leaderless transcription model similar to that of Gill-associated virus of Penaeus monodon prawns. Distribution of spliced reads between positive and negative strands mirrored that of non-spliced reads, that is most splices originated from positive strand reads. The proportion of reads from each strand that were spliced were similar, except in the case of de novo gill polyA selected libraries, in which low viral load samples (vCPM < 750), where the proportion of negative strand reads that were spliced was 2–3x higher than that of positive strand reads (supplemental figure SFig 1). However, given the low number of spliced reads and low total viral load in these samples, we are hesitant to draw any conclusions.

Figure 12. All splice sites identified in ribodepleted libraries.

Splicing events from all samples from newly generated libraries from the abdominal ganglion (A) and SRP136115 (B) are represented as arcs connecting two positions in the viral genome (x axis). Each splice pair that occurs more than once is rendered as a black arc, with the ten most common splices rendered as red arcs. Bars represent the sum of all donor (black) or acceptor (blue) sites within a 100bp window. A schematic of the Aplysia Abyssovirus genome is rendered above with major ORFs represented as labeled white boxes. Predicted subgenomic RNAs based on sequence homology to other nidoviruses are demarcated as colored regions [Trypsin-like_serine_protease (T, green), S-like_protein (S, cyan), Alphavirus_E1-like_protein (E1, yellow), Alphavirus_E2-like_protein (E2, gold), M-like protein (M, olive), and nucleocapsid-like proein (N, magenta)]. Major splicing events are primarily clustered within the first 5 kilobases and final 2.5 kilobases of the viral genome. Major splicing events are notably absent from the 5’ UTR (first 110 bases), the beginning of ORF2, and start of each predicted structural protein as would be predicted with coronavirus-like leader and body TRS splicing. Major splicing hotspots correspond to regions of increased coverage described in Figure 8.

Discussion

In this study, we confirmed the in vivo presence of the novel, apparently neurotropic Aplysia abyssovirus 1 first identified in silico by Debat 2018 (Bukhari et al., 2018) and investigated its transcriptional dynamics. Electron microscopy identified structures resembling DMV replication factories typical of nidoviruses in gill, suggesting infection and replication may be active in this tissue. Other structures resembling pleomorphic particles typical of coronaviruses suggest that AAbV may utilize a replication strategy similar to that of other nidoviruses such as the coronaviruses. Conclusive identification of these structures as unambiguously AAbV-derived would require more robust techniques, such as immune EM and/or true negative controls. If indeed these structures represent active replication in respiratory tissue, this would suggest that AAbV is a respiratory virus similar to the gill associated virus (GAV) found in shrimp (Cowley et al., 2002) and many nidoviruses of terrestrial vertebrates (Burrell et al., 2017; Hoon-Hanks et al., 2019; Tokarz et al., 2015). AAbV may then spread from active replication centers in the gill into other susceptible tissues, namely the nervous system and heart, as observed in other nidoviruses (Lavi et al., 1999). Although previously demonstrated to be related to nidoviruses such as coronaviruses (Bukhari et al., 2018), AAbV exhibited transcriptional features deviating from those typical of coronaviruses.

Due to the size of coronavirus genomes, efficient transcription of viral proteins requires discontinuous RNA synthesis of the viral genomes to generate sub-genomic RNAs (sgRNAs). This is achieved during negative-strand synthesis, via polymerases interacting with transcription regulatory sequences (TRS) located upstream of major ORFs, called body TRSs, mimic a TRS found near the 5’-end of the viral genome, called the leader TRS. Active coronavirus infection is thus typified by the presence of sgRNAs that can be readily identified by the presence of what appear to be high-frequency spliced reads that switch from copying the body TRS to copying the leader TRS, and creating a leader-body junction (Nomburg et al., 2020). Although putative, TRS leader and body sequence were identified in the AAbV genome, analysis of several datasets did not identify any evidence of a canonical set of splice sites or incorporation of a leader sequence that would suggest coronavirus-like use of leader-containing sgRNAs. Furthermore, the evidence either by RNAseq using polyA or ribodepeletion enrichment for enrichment of ORF2 relative to ORF1 by about 2-fold, was relatively modest in comparison to nearly 10-fold increases in read depth at each TRS in SARS-CoV-2 (Finkel et al., 2021). However, while the generation of sgRNAs is a derived characteristic of the order, the generation of several sgRNAs at manyfold higher levels than genomic RNA may be a synapomorphy of vertebrate nidoviruses that transcribe leader-containing sgRNAs.

An apparently basal member of the Nidovirales, GAV from shrimp Panaeus monodon, generates only two sgRNAs that are roughly as abundant as genomic RNA, and so would be expected to produce about a two-fold rise in read depth downstream of the body TRS (Cowley et al., 2002). The clade Pitovirinae within the Coronaviridae, known to infect fish and amphibians, have segmented genomes, with the structural proteins appearing on a different strand to the replicase, though it is not clear whether this represents a single sgRNA, or if further transcription of sgRNAs within the structural segment occurs (Edgar et al., 2022; Lauber et al., 2021). Integration of leader sequence into sgRNA is also not universal among nidoviruses. Toroviruses, such as the equine torovirus, are members of the Tobaniviridae which are closely related to the Coronaviridae, incorporate a leader sequence into some sgRNAs, and not others (Horzinek, 1999). Thus, AAbV appears to exhibit more basal nidovirus characteristics, i.e. generation of a single sgRNA and the absence of leader incorporation. The full range of transcriptional diversity within the Nidovirales remains unclear, and further study is required.

Although the leader-like sequence in AAbV is not spliced into sgRNAs as in more derived coronaviruses, the leader may facilitate sgRNA generation via a more basal mechanism. Like GAV, the leader and body TRS sequences in AAbV exhibit a high degree of complementarity, similar to but of greater magnitude than seen in coronaviruses (approx. 11 nucleotides in AAbV vs. 7 to 8 in coronaviruses). Stochastic annealing of negative and positive strand leader and body sequences during replication may act to stop the polymerase at the body sequence, thus generating ORF2 sgRNAs at stoichiometric ratios to full genomic RNA. Thus, TRS sequences facilitate generation of sgRNA without incorporation of any leader sequence via splicing.

In several nidoviruses, the 5’ UTR adjacent to the leader TRS sequence contain internal ribosomal entry sites (IRES). These sites recruit ribosomes for cap-independent translation after cap-dependent translation has been effectively shut off by virus-induced, stress response associated host translation shutdown (de Breyne et al., 2020). Thus, the leader and body TRS sites identified in AAbV may function as enhancers during viral hijacking of host stress response to facilitate viral translation at the expense of host gene translation.

Although canonical splice sites could not be identified, splice site analysis did identify a highly diverse landscape of non-canonical splicing events. Coronaviruses have similarly been observed to generate large amounts of noncanonical splices (Nomburg et al., 2020). Nomberg et al. 2020 postulate that these noncanonical splices in SARS-CoV2 are evidence of defective viral genomes (DVG). In our data, both PolyA selected and Ribodepleted libraries exhibit enrichment for the 3’ end of the AAbV genome, particularly adjacent to and including the poly-A region of the genome. The poly-A region of coronaviruses has been demonstrated to be essential for virus replication, even in the case of truncated DVGs (Spagnolo and Hogue, 2000). Several samples displayed high amounts of 5’-terminal and 3’-terminal sequence reads, suggesting the presence of DVGs, which are known to develop during infection with other nidoviruses (Girgis et al., 2022; Mendez et al., 1996; Molenkamp et al., 2000; Penzes et al., 1994).

The effects of DVGs on viral infection are idiosyncratic, but their production is known from diverse viral clades, including RNA viruses such as nidoviruses (Brian and Spaan, 1997; Genoyer and Lopez, 2019; Makino et al., 1988; Manzoni and Lopez, 2018; Rezelj et al., 2018; Vignuzzi and Lopez, 2019). DVGs are known to interfere with normal viral replication in some viruses and can lead to “semi-infectious” virions that replicate and spread but do not produce active infection (Genoyer and Lopez, 2019). In influenza, these semi-infectious virions can outnumber infectious virions 9 to 1 (Brooke et al., 2013). In coronaviruses, DVG are known to be capable of replication without a helper virus (Brian and Spaan, 1997).

In the murine hepatitis virus (MHV), a species of coronavirus, the DIssE sub-genomic sized DVG has been demonstrated to be able to replicate actively to the point of becoming the dominant RNA species in the host cell (Makino et al., 1988). Furthermore, persistent neural infection with MHV is associated with DVG replication and expression of spike and nucleocapsid proteins in absence of infectious particles. Notably, the nucleocapsid protein of coronaviruses is a key component for viral RNA replication, transcription, and translation and binds DVGs, even those lacking packaging signals (Cologna et al., 2000). Perhaps a similar phenomenon is responsible for AAbV comprising 20–40% of total RNA in some analyzed ribodepleted samples and enrichment of nucleocapsid RNA relative to other ORFs. In MHV, this phenomenon has been suggested to play a key role in the development of demyelinating encephalopathy and infection persistence (Adami et al., 1995).

Across viral taxa, DVGs have been suggested to play a key role in viral persistence (Genoyer and Lopez, 2019; Manzoni and Lopez, 2018; Rezelj et al., 2018; Vignuzzi and Lopez, 2019). In the case of paramyxovirus, high accumulation of DVGs in certain cell subpopulations leads to cellular survival via anti-apoptotic pathways and facilitate viral persistence (Xu et al., 2017). In hepatitis C, DVG have been speculated to act as decoys from the host immune system to allow for persistent infection as well as facilitating spreading efficiency (Karamichali et al., 2018). Given the evidence of AAbV as a persistent viral infection in Aplysia neurons, these putative DVGs may play a key role in viral persistence in Aplysia neural tissue as well. In the case of paramyxovirus, it has been suggested that cell specific host factors and potentially cell cycle state are determinants of DVG accumulation (Xu et al., 2017). Perhaps a similar phenomenon drives the differential accumulation of AAbV genome in diverse Aplysia tissues with notably extreme accumulation in neurons, which are in a permanently non-proliferative state.

In addition, the nervous system represents a unique suite of conditions that make viral clearance challenging. In general, the central nervous system is both non-regenerative and intrinsic to the proper function and survival of the organism. These features preclude the deployment of otherwise cytotoxic components of the immune response, hampering clearance of pathogens and infected cells. If the immune system fails to prevent neural infection, the neural immune response must resort to replication suppression and pathogen sequestration. Neurons are thus predisposed to persistent or latent infection (Liebert, 2001). Given Aplysia nervous system comprises relatively few neurons in simple neural circuits with little redundancy, the organismal cost of losing of any single neuron may be comparatively higher than in animals with much more neuron dense brains (Strumwasser, 1971). Given that the loss of even a single neuron is intolerable, Aplysia may permit the accumulation of viruses or viral nucleic acids to levels that may impair neuronal function to prevent outright loss of irreplicable components of neural circuits. This may explain why viral RNAs may accumulate in neural tissue as opposed to a tissue with high cellular turnover such as the gill where replication may be more active. The giant neurons of Aplysia have been documented to be extremely polyploid, perhaps the most polyploid of any somatic cell in the animal kingdom (Moroz, 2011a). This feature along with the unique immune considerations of neurons may explain why Aplysia neurons can tolerate extremely high loads of viral nucleic acids, upwards of 40% of non-rRNA species, without visible pathology. However, neuronal survival without obvious signs of pathology do not rule out sublethal impacts on nervous system function.

As previously mentioned, DVGs are a potent immunogenic stimulus even in the absence of infectious particle (Genoyer and Lopez, 2019; Manzoni and Lopez, 2018; Strahle et al., 2006; Sun et al., 2015). In influenza infection, the timing and magnitude of DVG accumulation has been demonstrated to play a role in virulence and infection outcome (Penn et al., 2022; Vasilijevic et al., 2017). Thus, even in the absence of an active infection, persistent infection typified by accumulation of AAbV DVGs in Aplysia neurons may still induces chronic inflammation as observed in MHV. Previous analysis of aging Aplysia neuron datasets used in this study highlighted the role of immune activation and inflammation in Aplysia neural aging suggesting AAbV may play a role in neural aging (Greer et al., 2018; Kron and Fieber, 2021). Further study is required to determine what impacts AAbV infection has on Aplysia, particularly the nervous system, given the apparently persistent nature of the infection.

An additional, unusual, phenomenon observed in the analysis of AAbV transcripts is the appearance of large proportions of negative strand viral transcripts in PolyA selected libraries in samples with relatively low numbers of viral transcripts. While negative sense genomic and subgenomic transcripts are an essential component of the viral life cycle of all nidoviruses (and indeed all positive stranded RNA viruses), these are usually present in small quantities relative to the positive strands (Alexandersen et al., 2020). The presence of these high proportions of negative strand AAbV RNAs in lower infection level samples was not tissue-specific as it was seen in subregions of two different ganglia (BSC and PVC) as well as in gill samples.

Only the positive sense RNAs, whether genomic or sub-genomic, should be polyadenylated. No sequences of contiguous adenosines in the AAbV genome longer than 6-A’s is found 5’ all but the last 1,000 bp of the positive sense genome. Thus, spurious capture of internal PolyA sequences seems unlikely. The occurrence of such negative sense transcripts in three different PolyA libraries from two different studies which employed different selection kits suggests that these results were not due to any technical failures in PolyA selection. Thus, the appearance of these negative sense AAbV transcripts in PolyA selected libraries remains a mystery. Determination of whether this represents an unusual stage of replication of this virus not appreciated in other nidoviruses or a novel type of DVG that has some polyadenylation will require further study.

Conclusions

The data presented here show that AAbV sequences are widespread in both hatchery reared and wild collected A. californica for at least the duration of time since this sequencing technology became available. Further, there is no reason to suspect that the virus was not always endemic in the wild populations, although perhaps variable in distribution and prevalence. Presumably the infection of hatchery animals originates with wild brood stock which are regularly introduced to the hatchery facility at the National Resource for Aplysia. The absence of high levels of sub-genomic relative to genomic RNAs as well as the lack of canonical splice sites for such sub-genomic transcripts suggests that AAbV is a relatively primitive nidovirus, more similar to GAV than the coronaviruses. Identification of the factors which maintain high levels of AAbV transcripts seen in many tissues but especially in neurons, perhaps throughout the adult lives of these animals, without any detectable loss of neurons will also require further study. That this virus was initially discovered only by mining RNA sequence databases and only after 50 years of intensive study of this species in hundreds or laboratories around the world suggests that obvious pathology is not associated with even high levels of AAbV transcripts in adult animals. However, researchers working with A. californica might wish to investigate associations between viral load and various gene expression, physiological and behavioral measures in their data. Further research is needed to conclusively tie observed virus-like structures with AAbV and to describe consequences of infection on the host.

Highlights.

• Virus present in wild and maricultured animals

• Viral replication only in gill but viral RNA accumulates in nervous system

• No apparent pathology despite high viral RNA load in nervous system

• Basal nidovirus that replicates with single sgRNA and no leader incorporation

• Viral RNA accumulating in nervous system likely defective viral genomes

Data Availability

Raw sequencing data is available at the NCBI Short Read Archive under BioProject PRJNA984706. Raw qPCR data is available in supplemental file SF3. All code used in the current study is available on GitHub in the following repository: [GITHUB REPO]