Low miRNA count - small RNA NGS

[kanrcc]

Hello to All,
I am a newcomer in this area and recently we initiated one small RNA NGS project to study the differential expression of miRNA in a cancer. We done (outsourced) small RNA NGS analysis using total RNA, isolated with mirVana kit, from 6 cancer cell lines . Library preparation was done using TruSeq Small RNA preparation kit V2 from Illumina according to manufacturer’s protocol. The small RNA library was selected using PAGE gel and the band between 145bp and 160bp was used for further sequencing. The selected libraries was sequenced in Single read 50 base run, using TruSeq SR cluster kit v3-cBot-HS and TruSeq SBS kit v3-HS, on a single lane of Illumina HiSeq1000 platform. The BaseCall files were converted into FASTQ files and demultiplexed using CASAVA v1.8.2. We obtained 31 to 43 million reads in each sample and %>=Q30 bases was around 96% in all samples. Then, we subjected each data file to adapter trimming and aligned to hg19 build refseq and small RNA annotations. The % of aligned reads was 80-83%. When we subjected to Genic Region filtering, the miRNA area was very low, only between 4 to 20% (read counts 1.7 to 7 million) in these samples. What may be the reason for such low count for miRNA in our NGS data? Is, it is a technical problem or data analysis (bioinformatics) issue? Also what will be the minimum % of miRNA area in total small RNA population for further valid analysis? I would be grateful, if you can help us to solve the problem.

[peterawe]

That seems very low indeed. It would be helpful to know what the aligned/non-miRNA reads are. If it's abundant rRNA/tRNA it indicates degraded input/RNA sample. For my cancer cell lines samples, 20-80% of aligned reads are mapped to miRNAs. 20% is the exception and have been due to old/partially degraded RNA. I would check the following (in prioritized order): 1. Be sure you are not counting collapsed reads 2. Find out what your-non miRNA reads are 3. RNA quality of your input (Bioanalyzer/RIN) That being said, 1.7 - 7 million miRNA/sample are in general very good for most applications and the data may still be adequate depending on your question and proper filtering. Good look with the troubleshooting!

[dlawrence]

The % is low but not out of the ordinary. The relative proportions of different small RNA classes can change drastically between experiments. As peterawe says, usually it is rRNA and tRNA but I have also seen almost pure snoRNA - we called that sample 'the blizzard'

This paper (Reproducibility of high-throughput mRNA and small RNA sequencing across laboratories) is pretty good:

http://www.nature.com/nbt/journal/v31/n11/full/nbt.2702.html

Particularly useful is: "differences in the proportions of the different sRNAs are likely introduced during RNA extraction"

Good luck!

[peterawe]

That paper is excellent! In particular I find the figure where they delineates the relative importance of sources of variation in RNAseq very interesting. If I remember correctly, they also states that the percentage of t/rRNA reads has surprisingly low impact on your miRNA expression analysis. I liked your blizzard term... Did you find out what caused it?

[kanrcc]

Thank you for the immediate responses. I have used the Avadis NGS package for the analysis. In its Genic Region pie diagram, major part depicted as 'others' which may include piRNA, rRNA, etc and then comes intronic reads. These two together forms 70-80% of the total reads. SnoRNAs and tRNAs are only a small portion. Then reads are not collapsed ones, clean reads only. All RNA samples have RIN 9.7-9.9. The aim of the present study is to pick out the differentially expressed miRNAs in these cancers. Hence, whether this wide variation in the count will influence the result.

[peterawe]

Dlawrence ref adress this issue in some detail: "In small-RNA sequencing, the microRNA (miRNA) content differed widely between samples owing to competitive sequencing of rRNA fragments. This did not affect relative quantification of miRNAs." It's a very good read.