Hello Amruta,

If the mRNA is polycistronic, you'll be able to find multiple termination codons and multiple ribosome attachment sites. Thus you can identify the genes involved in an operon.

I have some unreleased software for identifying "transcription units" (TUs) based on strand-specific (!) RNA-Seq. I can run it for you - it was designed for bacteria, but will definitely handle any quantity of reads within reasonable time. However, it does not understand exons/introns. It can be applied to strand-nonspecific data, too, but at least for bacteria I advise against doing so. Also, sequencing data quality (mostly in terms of the absence of spatial bias) may limit the ability to perform operon identification automatically.

Specifically for TUs, I'm detecting them based on:

- distance between the two candidate CDS annotations (one expects a full bacterial promoter to be 35-40bp long, although promoters can also be "hidden" within upstream genes; this setting is flexible, and can be set to 0bp);

- continuous coverage between the two CDS, with user-specified minimal allowed coverage;

- each CDS pair is evaluated separately, which leads to a gradual elongation (or not) of a potential TU.

When you have a specific list of genes possibly forming an operon, I'd very strongly recommend examining coverage "manually" (in a genome browser). This will at the very least protect you from unnoticed RNA sequencing artifacts. When examining, pay attention to GC content of separate segments - this may affect coverage in a fairly predictable manner :)

As a first test, you could try to run standard transcript reconstruction tools, such as Cufflinks or Trinity (see links below), and compare the inferred transcript structures to annotated coding sequences (or ORF predictions for the transcript assemblies). For strongly expressed policistronic RNA, you expect to see reads spanning regions with multiple coding frames. In the transcript reconstruction output, you could thus look for things like "merged transcripts", which are probably good candidates for operons (although some errors that Cufflinks and related tools make, e.g. merging genes that are very close to each other, could look similar).

http://cufflinks.cbcb.umd.edu/

http://trinityrnaseq.sourceforge.net/

Hi Amruta,

As you already know that the eukaryotic operons, especially those in nemtaodes are different from the canonical bacterial operons, and work via trans-splicing, their identification in nematodes requires customized approaches.

We have recently published a paper tackling exactly the same problem - how to identify operon like gene organization in another model nematode Pristionchus pacificus!  Here is the web-link and the citation for our work (also attached below):

http://dx.doi.org/10.1261/rna.041954.113

Sinha, A, Claudia L, Sommer RJ, and Dieterich C. 2014. “Genome-wide analysis of trans-splicing in the nematode Pristionchus pacificus unravels conserved gene functions for germline and dauer development in divergent operons.” RNA. doi:10.1261/rna.041954.113.

Briefly, we first carried out a pull-down on mRNA using oligo-probes complementary to either the SL1 and SL2 spliced leaders present at the 5-prime ends, and then carried out RNAseq on these two fractions to find which transcripts get which spliced leader. We also computationally predicted operon-like clusters by searching the genome for closely spaced genes in the same orientation. We then combined the RNAseq results with these computational predictions to identify experimentally validated operons in P. pacificus.

But if you already have some RNAseq data and you do not wish to do the mRNA pull-downs, you may use an alternative approach such as that used in Allen et al 2010:

Allen, MA, Hillier LW, Waterston RH, and Blumenthal T. 2010. “A Global Analysis of C. elegans trans-splicing.” Genome Research. doi:10.1101/gr.113811.110.

In this approach, if you have enough sequencing depth, you expect to recover the 5-prime ends of various transcripts, from which you could then count the instances of SL1 or SL2 sequences observed at the 5-prime end of the transcript. The only catch that I see here is that 5-prime ends tend to be poorly represented in RNAseq data and so the data could be a bit noisy. We avoided this in our experiments by first starting with a SL-specific pull-down on the 5-prime end, before doing any RNAseq experiments.

You may also try the approach outlined above by Georg, although in my experience as well as based on reports from other groups (e.g. Blumenthal et al), the poly-cistronic transcripts usually get resolved into individual cistrons quite fast, making them much harder to detect by RT-PCR or RNAseq.

I hope this is useful. If you have any other questions, please let me know.

Best,

-Amit.

http://dx.doi.org/10.1261/rna.041954.113

http://dx.doi.org/10.1101/gr.113811.110

Thank you so much Dr. Amit. I will definitely look into those papers. 

Also, thank you all for the great ideas.