incremental strategy for RNA expression profiling

[ChristmasSunflower]

I have one technical question regarding to gene expression profiling and I like to consult with the expert here, to see how you manage this as Director in a sequencing center or core facility.

As you can understand, from the point of a sequencing center or core facility, we are dealing with different set of samples from different PIs. I can understand PIs want to have their sequencing data as a whole set ready within single run. For instance, one RNA-Seq project has 10 RNA samples. PI wants to use 2 lanes of 50SR for just regular gene expression study. So the pooling scheme will be 5 samples per lane and we target to generate about 30M reads per sample, if we say one lane produces about 150M reads. As you know, this is very common to have sample reads variation in the pooling. For this 10 RNA sample project, we generate 60M reads for sample 1 and only 15M reads for sample 2. At my position, I will say I can generate another 15M reads for sample 2 in our next run either using the same library if we still have it available or make new library to make up 15M reads. So PI will have 30 M reads combined together for sample 2 for data analysis. I heard this kind of incremental strategy may be NOT appropriate for RNA expression profiling. I'm not sure how you manage this at your place. For RNA-Seq or small RNA-Seq these kind of gene expression profiling experiment, how do you deal with this case, adding 15M reads for sample 2 or you just re-run sample 2 with the target of 30M reads in the next run? Or you do some analysis (for instance R2 correlation calculation) to make sure you can combine reads from different runs?

This is very important issue for me to think about. If this is true, it could significantly change our experimental set up-I will have to ask PIs to sequence more in the first time to try to avoid any make up later. Also based on your experience and knowledge, as well as some observations from different sequencing centers and facilities, what are your comments/thoughts on this?


I will appreciate your thoughts on this very much.

Thanks
ChristmasSunflower

[pbluescript]

It is just another variable added to the experiment that you will have to properly control for. I would recommend not doing it unless you had some sort of spike in control that you could use to normalize samples sequenced on multiple flow cells. The ideal solution would to be sure your library quantification is as accurate as possible so you get as little library imbalance as possible. qPCR is a very good way to do that. It won't eliminate it, but it should help.
Regardless, for gene level expression studies, 15M reads is usually good enough as long as most of them map uniquely.

[ChristmasSunflower]

Thank you very much for sharing your thoughts, pbluescript. I totally agree with you the ideal situation is to quantify library accurate as much as we can. We use combination of Q-PCR, Qubit and bioanalyzer. But sometimes, we will have to face the reality of read number variation across the samples in a project. The issue is getting complicated in small RNA-Seq library quantification. As Illumina small RNA protocol says mix equal volume before running gel to separate the band, we tried to use bioanalyzer peak quantification way to specific quantify small RNA peak to hope to lead better pooling. Sometimes it worked well but sometime not.

What I'm trying to ask advice here is usually how sequencing center or core facility deals with this variation just like what I wrote in my first post in this thread. Also i'm not sure what you meant "some sort of spike in control that you could use to normalize samples sequenced on multiple flow cells". Suppose sequencing follows the poisson distribution and this incremental strategy could be applicable. But I agree we need to be careful, for sure.

Thanks

[Genohub]

pooling samples

I recommend barcoding each sample at the ligation step, careful quantitation of samples by qPCR, normalizing to the lowest concentration and then pooling all your samples in the same tube. If the customer wants two lanes, distribute all samples across both lanes (assuming they want an even number of reads per sample). This should limit lane effects across your samples.

Finally I would recommend telling users that the number of reads they can expect depends on their library quality and will certainly be different from sample to sample. Instrument specs are one thing, but in reality, YMMV.

See the pooling recommendation in this blog post for more details.