Ribosomal RNA contamination assessment

[ilivyatan]

I have data from an RNA-seq experiment on the SOLiD platform.
One round of Ribominus was used before library preparation to get rid if rRNA.
I want to assess how well this process worked.
Using bowtie I made an index of all the rRNA sequences I could find (in mouse) and aligned against it. Result was 10-11% of my reads mapped to rRNA.
I also mapped my reads to the entire genome obtained in colorspace format from bowtie's website.
My question is does the genome index include the rRNA sequences? Should I subtract the 10-11% from my mapped reads?

Thanks.

[malachig]

Out of curiosity, where did you obtain your list of rRNA sequences from? Since rRNA genes are genes like any other it seems likely that they would be contained within the mouse genome sequence... Why is is necessary to subtract them from your mapped reads?

You may find that even after ribominus reduction that a fair amount of ribosomal sequences (and therefore reads corresponding to them) remain, but their presence is not necessarily detrimental to downstream analysis. If the goals of your analysis are to compare expression between conditions, tissues, etc. then I would say that it is more critical that the ribominus worked equally well for all samples...

The issue of how to deal with Ribosomal sequences in RNA-seq data is an interesting one and has been discussed in the forum before. For example, here:
Is it a good choice to filter out reads using repeatmask?

[NextGenSeq]

The best way to do this is before you sequence it and to run a Bioanalyzer chip of the before and after Ribominus RNA. That way you won't waste library and sequencing reagents if you have too much rRNA contamination.

[malachig]

Good suggestion. We used to use a RiboMinus kit on total RNA in advance of hybridization to Affymetrix exon arrays. We would then run a Bioanalyzer chip and if the rRNA contamination was still too high we would either start again with fresh total RNA or perform a secondary RiboMinus step on the same RNA (i.e. a double reduction).

[ilivyatan]

rRNA contamination

To answer a few questions...
The rRNA sequences used to create the index were manually downloaded by me via the Entrez nucleotide database. There aren't too many so this is not difficult.
Second, I performed one round of Ribominus and analyzed with Bioanalyzer before sending for sequencing as you suggested but I'm still interested in assessing the amount of contamination still left (for reasons I can't disclose now!).
That's why I want to do an after-the-fact computer analysis of contamination.
I'm not sure the assumption that the ribosomal sequences are in the genome is correct since at least one (28s) blasts to chrUn (unassembled chromosome) and seems not to be included in the genome assembly as far as I can tell.
Anybody have more info?

[scooter]

Have you tried Ribo-Zero from Epicentre? I hear it works much better than Ribominus....

[malachig]

I'm not sure the assumption that the ribosomal sequences are in the genome is correct since at least one (28s) blasts to chrUn (unassembled chromosome) and seems not to be included in the genome assembly as far as I can tell.

Interesting point. The complete 'genome build' consists of both those things that can be assigned to a chromosome as well as those things that have not yet been assigned to a chromosome. The human genome build has reached a highly finished status but it still has contigs that have not yet been placed. These are still part of the build and may contain important sequences including genes. In other species with less polished genomes, the number of contigs is much larger than the number of chromosomes for the species (in some cases several orders of magnitude higher). For this reason, when aligning reads to a genome you should probably be aligning to the complete genome build not just the named chromosome contigs.

For human, the number of contigs has stabilized but gradually decreased over the last several builds. Stats here for: build 35 (hg17), build 36 (hg18), and build 37 (hg19).

NCBI summary notes for all human releases.

NCBI describes is like this:
"reference assembly: RefSeq records representing the official reference genome assembly include sequences assembled into chromosomes 1 through Y (RefSeq accessions NC_000001 through NC_000024), the human mitochondrion, and RefSeqs representing sequences that were not localized on a chromosome."

[malachig]

The latest issue of Nature Methods includes an article on Validation of two ribosomal RNA removal methods for microbial metatranscriptomics. Although involving prokaryotic species, their methods for assessing efficiency of rRNA removal may be helpful. The same issue contains an application note on the Ribo-Zero rRNA removal method mentioned by scooter. They have a kit for human/mouse/rat and another for bacteria.

[pmiguel]

The latest issue of Nature Methods includes an article on Validation of two ribosomal RNA removal methods for microbial metatranscriptomics. Although involving prokaryotic species, their methods for assessing efficiency of rRNA removal may be helpful.

Thanks! Most of the information is what one might expect, but nice to have it nailed down. One interesting point that had not occurred to me:

Supplementary Note 4. Difference between actual and observed rRNA removal
The observed post-depletion rRNA fraction does not accurately reflect the amount of rRNA that is actually removed. To illustrate this, let us assume a hypothetical total RNA sample containing 95 rRNA and 5 mRNA molecules (i.e. 95% observed rRNA). Removal of 80% of the rRNA leaves 19 rRNA molecules and the original 5 mRNAs (i.e. 79% observed rRNA). This demonstrates that the observed rRNA depletion is only 16% despite 80% actual removal.

Get that? If you are starting with 95% ribosomal RNA, you would need to remove nearly 99% just to drop the % of rRNA reads down to 17%!

--
Phillip

[jscaria]

The latest issue of Nature Methods includes an article on Validation of two ribosomal RNA removal methods for microbial metatranscriptomics. Although involving prokaryotic species, their methods for assessing efficiency of rRNA removal may be helpful. The same issue contains an application note on the Ribo-Zero rRNA removal method mentioned by scooter. They have a kit for human/mouse/rat and another for bacteria.

This reply may be bit late. However thought that the info may be useful for those might look up this later.

I am using the Epicenter Biotech product Ribo-zero and it works well. You can remove 90-95% rRNA with this kit.

[carmeyeii]

pmigue,

that is a very good point!

[carmeyeii]

I am analyzing some Illumina libraries that appear to have a lot of ribosomal RNA contamination.

I'm using Bowtie to align the reads only to a specific set of sequences, and because of the differing amount of rRNA contamination in each sample, each of them maps a different percentage of reads to the dataset (some half of what others map), ranging from 1% to 0.3%.

I wonder if the amount of rRNA contamination in the preparation of a library can have an impact on the apparent expression level of a gene -- even though one normalizes its counts agains the total number of reads that mapped.

What's your opinion in this subject?

Carmen

[puggie]

I know this is an older thread, but Im thinking about this issue during my data anlysis. Lets say one has two data sets A and B: rRNA depletion was 50% for A and 75% for B. Then if rRNA reads are not filtered at some point, e.g. prior to normalization when taking library size into account, then the estimate would be influenced by an "external" factor which is the step of rRNA removal prior to sequencing. E.g. if library size appears increased due to reads mapping to ribosomal sequence.

[puggie]

Update: I tried now to map against rRNA "assembly" this afternoon and got between 8-40% of total reads for 5 samples mapping to rRNA. These are degraded/old freezer samples subjected to NGS, so I was expecting rRNA depletion (RiboZero) not to remove non-intact rRNA sequence. The samples are murine. Annotation used: gi|38176281|tpg|BK000964.1| TPA_exp: Mus musculus ribosomal DNA, complete repeating unit