Efficient subtraction of insect rRNA prior to transcriptome analysis of Wolbachia-Drosophila lateral gene transfer

Background Numerous methods exist for enriching bacterial or mammalian mRNA prior to transcriptome experiments. Yet there persists a need for methods to enrich for mRNA in non-mammalian animal systems. For example, insects contain many important and interesting obligate intracellular bacteria, including endosymbionts and vector-borne pathogens. Such obligate intracellular bacteria are difficult to study by traditional methods. Therefore, genomics has greatly increased our understanding of these bacteria. Efficient subtraction methods are needed for removing both bacteria and insect rRNA in these systems to enable transcriptome-based studies. Findings A method is described that efficiently removes >95% of insect rRNA from total RNA samples, as determined by microfluidics and transcriptome sequencing. This subtraction yielded a 6.2-fold increase in mRNA abundance. Such a host rRNA-depletion strategy, in combination with bacterial rRNA depletion, is necessary to analyze transcription of obligate intracellular bacteria. Here, transcripts were identified that arise from a lateral gene transfer of an entire Wolbachia bacterial genome into a Drosophila ananassae chromosome. In this case, an rRNA depletion strategy is preferred over polyA-based enrichment since transcripts arising from bacteria-to-animal lateral gene transfer may not be poly-adenylated. Conclusions This enrichment method yields a significant increase in mRNA abundance when poly-A selection is not suitable. It can be used in combination with bacterial rRNA subtraction to enable experiments to simultaneously measure bacteria and insect mRNA in vector and endosymbiont biology experiments.


Background
Many interesting bacteria form intimate, obligate relationships with eukaryotes. These bacteria include endosymbionts and obligate intracellular pathogens. These microbes can be difficult to research as they cannot be cultured, easily manipulated, or genetically transformed. Therefore, genomics techniques have significantly advanced the study of these organisms. Because of the obvious potential health impacts, many techniques and tools have been developed to research such bacteria that interact with humans. For example, the Ribo-Zero rRNA removal kit for human/mouse/rat (Epicentre, Madison, WI, USA) can facilitate transcriptome analysis of intracellular pathogens of humans. However, for non-mammalian systems including bacteria/vector systems, a void still exists. Previously, the MICROBEnrich insect/C. elegans module (Ambion, Austin, TX, USA) was used for this purpose, but it is no longer available. Therefore, we sought to investigate if Epicentre's Ribo-Zero rRNA removal kit designed for humans and rodents would efficiently remove rRNA from insect samples. Drosophila ananassae is a fruit fly that can be naturally infected with a Wolbachia endosymbiont [1,2]. In addition, some Wolbachia-colonized lines have an entire Wolbachia genome transferred to a fly nuclear chromosome [3]. Transcripts from nuclear Wolbachia transfers (nuwts) are of particular interest, as they can be used to elucidate functions for the nuwts. Previously, we identified 28 nuwt transcripts (~2% of the Wolbachia genome) in adult flies, albeit at low levels [3]. Since these transcripts might not be polyadenylated, poly-A enrichment of total RNA is not a suitable technique for obtaining mRNA prior to RNASeq. Therefore, Ribo-Zero was tested as a suitable alternative.

RNA isolation, mRNA enrichment, and transcriptome sequencing
Total RNA was isolated with TRIzol, as previously described [3], from 50-85 freshly laid eggs of wild-type D. ananassae In(3R)A (Stock No. 14024-0371.13) that is infected with the Wolbachia endosymbiont wAna [2] and from a tetracycline-treated (Wolbachia-cured) line of this fly. Both lines are tested regularly by fluorescence in situ hybridization to confirm the presence or absence of a Wolbachia infection, respectively. The samples were enriched for mRNA using the Ribo-Zero rRNA removal kit for human/mouse/rat (Epicentre, Madison, WI, USA), following the manufacturer's protocol.

Assessment of rRNA removal by microfluidics
Comparison of the original and Ribo-Zero-subtracted samples on a Bioanalyzer (Agilent, Santa Clara, CA, USA) illustrates the efficient removal of the majority of the rRNA (Figure 1) and the loss of >97% of the RNA, as assessed by integrating the area under the rRNA peaks. In D. ananassae, as well as in many insects, the 28S rRNA is naturally cleaved resulting in two peaks that are unusually close together on the Bioanalyzer when compared to other eukaryotic or bacterial rRNA samples. Importantly, the Ribo-Zero subtraction efficiently removed both halves of the cleaved rRNA.

Assessment of rRNA removal by transcriptome sequencing
While the Bioanalyzer is highly sensitive at detecting intact rRNA, degraded rRNA can go undetected. Therefore,  Figure 1 Bioanalyzer analysis of Ribo-Zero-subtracted RNA. The subtraction of rRNA using the Ribo-Zero human/mouse/rat reagents was tested on total RNA from Drosophila ananassae with (panels A and C) and without (panels B and D) the presence of its Wolbachia endosymbiont. The total RNA prior to Ribo-Zero subtraction (panels A and B) is compared to RNA after Ribo-Zero subtraction (panels C and D). The starting amount of RNA prior to subtraction for panels C and D was equivalent to the amount shown in panels A and B. Ribo-Zero-subtracted RNA Total RNA Figure 2 Comparison of total RNA and Ribo-Zero-subtracted RNA from Drosophila ananassae. The reads from both samples were mapped with BWA against the D. ananassae scaffolds. Each scaffold was then computationally divided into 100 kbp fragments and the number of reads mapping to the fragment were counted. When scaffolds were <100 kbp, the entire scaffold was counted. The last fragment of each scaffold was always <100 kbp. Fragments containing annotated rRNA are shifted right of the diagonal (gray line) due to decreased representation in the Ribo-Zero-treated RNA. Meanwhile, fragments without rRNA are shifted leftward of the diagonal (gray line) because of their increased abundance relative to the entire sequenced population in the subtracted samples.
we also sought to examine RNASeq results to ensure even subtraction across the rRNA. Paired-end libraries (Illumina, San Diego, CA, USA) were constructed using the two pools of RNA for the cured line ( Figure 1B and 1D) using the standard protocol starting immediately after the poly-A selection step. Half of a channel of 72-bp reads was obtained on a GAIIx (Illumina, San Diego, CA, USA) for each library. The sequencing reads were mapped against the reference D. ananassae assembly, using the default parameters for BWA [4], yielding 15.8 million and 15.0 million mapped reads from total RNA and the Ribo-Zero-subtracted sample, respectively. The reads for each 100 kbp window across the genome are plotted, comparing the subtracted sample from the total RNA ( Figure 2). In this subtraction, and unlike the subtraction with the RNA from uncured specimens, the Bioanalyzer revealed that only 98% of rRNA was removed. Therefore, and as expected, a significant amount of rRNA is still present, as illustrated by the red dots representing fragments with at least a portion of an rRNA. However, the results are consistent with removal of >90% of the rRNA with a shift leftward of the majority of 100 kbp fragments due to enrichment for mRNA and separation of the rRNA from the mRNA by approximately one order of magnitude. While the overall trend is clear, specific points may not reflect this trend. For example, the 100 kbp region containing the 18S rRNA is the red point in the upper right corner that appears unshifted from the diagonal. In the total RNA, 2,936,794 reads mapped to the 18S rRNA fragment (41,477-43,516 bp on scaffold 13163; GenBank CH902719.1); in the subtracted sample 980,094 reads (66%) were sequenced ( Figure 3A). This would be sufficient to offset the point from diagonal except that an additional~3 million reads map in this 100 kbp fragment and outside the 18S rRNA. These reads prevent the full shift of this point relative to the diagonal that would be realized should only the 18S rRNA be present in this fragment.
While there is a 66% difference in the raw number of 18S rRNA reads between the two samples ( Figure 3A), this does not fully capture the Ribo-Zero subtraction. Subtraction of rRNA means that significantly more reads are obtained from the remaining RNA species in the subtracted sample, as seen in the 6.2-fold increase in signal for actin in the subtracted sample ( Figure 3C). When the rRNA results for each sample are normalized to the  The coverage of reads across the 18S rRNA (Panel A) and actin (Panel C) genes was determined by mpileup in samtools and compared between total RNA (solid line) and Ribo-Zero-subtracted RNA (dashed line). For the 18S rRNA, the Ribo-Zero-subtracted sample contained 66% fewer rRNA reads when compared to the total RNA sample. When the 18S rRNA results are normalized relative to actin, a 95% reduction in the rRNA is seen (Panel B). The 18S rRNA is only partially sequenced in the reference genome with a gap in the scaffold to the immediate right of this region.
number of reads in actin, a 95% reduction of rRNA is observed ( Figure 3B). This is consistent with the 98% reduction determined by integrating the area under the peak on the Bioanalyzer.
Not only does the subtraction increase the signal for genes as shown above with actin, it also increases the number of genes that can be analyzed. While 8,888 transcripts had at least a single read mapping in the subtracted sample, only 7,629 transcripts had a single read mapping in the unsubtracted sample. Yet, a single read is not very informative when examining differential expression, and instead, a minimum number of reads/transcript may be required for a differential expression analysis. Standards for this minimum have not been established to our knowledge. But if one required 100 reads/transcript, the subtracted sample would have 3,677 transcripts that could be analyzed while the unsubtracted sample would only have 1,047 transcripts.

Detection of transcripts arising from bacteria-to-animal lateral gene transfer
While the rRNA was sufficiently removed from the studied samples, only a few reads (19 and 20 reads from the total RNA and the Ribo-Zero-subtracted RNA, respectively) arose from transcripts of nuwts as identified by mapping with BWA [4] against the wRi reference genome [5] (Table 1). No transcripts were identified with more than one mapped read after duplicate removal. This low abundance of reads is not sufficient for analysis of the nuwt transcriptome. There was no overlap in the reads found between the two samples with the exception of reads that likely arose from non-Wolbachia bacterial contaminants on the surface of the eggs, further suggesting the stochastic detection of these low abundance transcripts.
This low abundance of transcription mirrors previous findings that transcription of nuwts is low [3]. In previous work, nuwts were found to be 10 4 -10 7 times less abundant than actin in adult flies [3], which is consistent with the RNASeq results presented here for eggs. This level of transcription may or may not be biologically relevant. Important tissue-, condition-, and/or stage-specific transcription cannot be ruled out. However the tissue, condition, and/or stage that should be examined are not immediately obvious in the absence of a phenotype.

Conclusions
The Ribo-Zero rRNA removal kit for human/mouse/rat efficiently removes >98% of insect rRNA from total