Uneven reads in pooled sample

[theduke]

Hi All,

I've been doing this a while, but have run into a new problem on a batch of RNA-Seq samples (Truseq) that I'm concerned about. We prepped a bunch of RNA samples for sequencing and have a few that are low concentration. The core facility that I have been using for years was unhappy with this, but I didn't really see where the issue was given there was plenty of material for pooling.

Nevertheless they ran one lane for me (HiSeq), but apparently had to modify their dilution protocol. The result was as follows for the lane:


Reads % lane % PF cluster Qual score
Sample*1 31,287,035 12.66 89.81 36.96
Sample*2 70,103,756 28.36 89.49 36.96
Sample*3 8,599,888 3.48 90.17 36.97
Sample*4 3,309,119 1.34 89.80 37.01
Sample*5 98,775,103 39.95 89.24 36.94
Sample*6 28,415,646 11.49 90.10 36.94

Now, clearly the issue is that the distribution of data is very uneven, BUT the number of reads directly and perfectly correlate with the starting concentration of each sample prior to pooling (R2 = 0.99)! From what I can gather, the core did not make an intermediate dilution for each but instead added variable volumes to a pool and used that to cluster, but what happened in between I do not know. What I dont understand is why they did this. Here are the sample concs:

ng/ul nM
Sample1 1.6 6.3
Sample2 2.4 9.6
Sample3 0.9 3.5
Sample4 0.6 2.4
Sample5 2.7 10.8
Sample6 1.5 6.1

We are almost certain that given the correlation with the pre-pooled sample concentration the pooling has been done incorrectly and that the variable volumes has perhaps thrown everything off balance. Does that sound reasonable?

According to all the illumina documentation I have read, the best thing to do would be to normalize each sample to 2 nM, and then combine equal volume of each together to make a 2 nM pool, which would then be the starting point for cluster generation. Does this sound reasonable? Am I missing something here?

****I should mention that samples were quantified with both picogreen and bioanalyzer HS DNA. These gave very similar results (r = 0.8) so I doubt quantification is a major issue here****

[nucacidhunter]

If the library sizes were similar, they have failed to do equimolar pooling. Using different volumes for pooling or equal volumetric pooling of normalised libraries should not affect the read ratios if done correctly. qPCR is the most accurate library quantification method and the ones mentioned here are not.

[theduke]

Agreed that qPCR is the way to go for accurate quantification, but that's a separate issue. Here we are talking about relative concentrations and the pooling of multiple samples. It certainly looks like equimolar pooling has failed in this instance.

[HESmith]

We've consistently observed lower-than-calculated cluster numbers when working with dilute (<1ng/ul) libraries. We don't think it's an issue with adsorption b/c we use LoBind tubes, nor an issue with the quantification (using PicoGreen fluorometry and BioAnalyzer, same as OP). We suspect that it's related to library quality, particularly when prepared in parallel. If the same amount of RNA input yields substantially different amounts of library, it's reasonable to assume that something wonky happened.

Also, you may want to recalculate your correlation. For example, sample 2 (9.6 nM) is 4X the concentration of sample 4 (2.4 nM), but produces 21X as many reads (70M / 3.3M).

[theduke]

What was the cutoff for the observed weirdness? <1ng/ul? I think we will redo the 5 or 6 samples that are odd ones out and start over.

Does my pooling plan sound reasonable?

[nucacidhunter]

We've consistently observed lower-than-calculated cluster numbers when working with dilute (<1ng/ul) libraries. We don't think it's an issue with adsorption b/c we use LoBind tubes, nor an issue with the quantification (using PicoGreen fluorometry and BioAnalyzer, same as OP). We suspect that it's related to library quality, particularly when prepared in parallel. If the same amount of RNA input yields substantially different amounts of library, it's reasonable to assume that something wonky happened.

Also, you may want to recalculate your correlation. For example, sample 2 (9.6 nM) is 4X the concentration of sample 4 (2.4 nM), but produces 21X as many reads (70M / 3.3M).

Lower than expected yield indicates some issues with library prep but it could be unrelated to quality, for instance loss during library clean-up of a good library. If there was an issue with library quality, then the cluster density and yield should be lower than expected as the bad library would not contribute to sequencing output. Because of formatting issue I can't figure out the stats given on output here. Pooling equimolar quantity of libraries quantified with qPCR with below 1 ng/ul and up to 10 ng/ul I have not observed this.

[theduke]

Because of formatting issue I can't figure out the stats given on output here.

Apologies for that. In the first table, there are five columns in the following order:

Sample name
Number of reads
% of the lane
%PF clusters
Quality score

There doesn't appear to be any quality issues per se.

[nucacidhunter]

Sequencing reads is over 230M which is around average for V4 chemistry on HiSeq 2500 if that is the platform used for sequencing and read quality and PF is up to specs.