library construction from pooled individuals

[coutellec]

Hi everyone,

We would like to test population effect on DE (face to a toxicant) using mRNAseq. We need samples to be representative of each population.
Is it correct to pool different individuals to construct a library, and to consider libraries made from different sets of individuals as biological replicates at the population level ? (e.g. 1+2+3 = R1; 4+5+6 = R2). Do you know exemples where such libraries have been used ?
Thanks in advance for any advice

[Simon Anders]

Can you explain in more detail what you want to do? What do you mean by "population effect on DE"?

[coutellec]

Dear Dr Anders,

Thanks for your interest.
By "population effect on DE", I mean we want to compare the transcriptomic response to a model chemical (constitutive vs chemical-induced expression, using control and experimentally exposed individuals), across different populations of a freshwater invertebrate. Populations (14 in total) are geographically distant and occupy sites with contrasted environmental characteristics. We have sampled adults in the field and lab-reared a F1 from 12 families per population. These F1 indiv have been subjected to the chemical exposure experiment. Using F1 allows reducing environmental influence on trait expression, and we thus consider variation among populations to be genetic (see Fst- Qst approaches).

As we are interested in the "between-population" level of variation in gene expression, we want to have biological replicates that are representative of this level. Idealistically, as in a quantitative genetics approach, we should replicate families within populations (each replicate = one distinct famly) and even replicate individuals within families (intra-family variation). However, due to financial constraints, we cannot afford such a design. To increase population representativeness, we thus plan to pool individuals from different sets of families to build our biological replicates. WIth regard to the chemical, we might consider we dispose of a high number of biological replicates (n pools of family-sets times p populations).
Gene expression is to be quantified with RNA-seq and Illumina Hiseq2000 data. Each library should be barcoded. We have the money for 1 flow-cell, so that we should limit our design to 20 libraries.
Hoping it helps understand, and sorry if this message was posted several times (internet pb).
Regards,
Marie-Agnès

[Simon Anders]

I think your design make sense, but I am puzzled why you think you can only have 20 libraries on one flow cell. People have used up to 96 barcodes in a flowcell lane, so you can have much more, unless library preparation becomes to expensive. (Remember that your total cost for sequencing one flow-cell is fixed, and with multiplexing, you can put many more samples in than there are lanes. Of course, you need to budget for each library the cost of sample prep until the stafe where you have the multiplex tags attached and can pool everything.)

Pooling several individuals from a population is certainly a good compromise. It is important, though, that you represent each population (for both the treated and the control condition) with several pools. Only then can you compare pools representing the same population treated the same way with each other to see how much variance remained that the pooling could not rid you of.

Hence, what I would do is to have several pools for each population, each pool comprising individuals from each family (but no individual mixed into more than one of the pools). The within-group variance is then the wthin-population-between-family variation and the within-family variation, attenuated by the pooling gain, and this what you need to compare between populations.

There is a common misconception that confuses people, so to quickly mention this: If you have the choice of sequencing either two replicates to 10 million reads per sample or four replicates to 5 million reads per sample, you should always go for the latter. Your power depends on the total number of reads per group, not on the number of reads per sample, and sequencing costs depend on the latter, too.

[coutellec]

Thanks for your answers et advices. Effectively, I considered samples as conditions. Then we should revise our design with more samples, which is really good news.

[dbroh11]

I know this post is fairly old, but in a similar vein, we are interested in examining the effects of four different drug compounds (with 1 being a "control" in the form of a growth factor) on differential gene expression in a motor neuron cell line.

We are planning to split the same motor neuron cell culture into 3 separate dishes for all 4 drug exposure conditions (12 total dishes per experiment). We would then isolate RNA from all three dishes per drug, combine it all and add a single unique barcode as it is taken through Total Stranded RNA-Seq prep for Illumina sequencing.

After reading many Anders posts, I see the inherent problem if we were to only test 4 total barcoded samples against each other (even though each is being comprised of RNA from different treated cell cultures) as its essentially one replicate per condition.

Would the solution then be to repeat this experimental design at three different time points, and then run the 12 barcoded samples (each representing aggregate RNA from three split cell cultures to a given drug response at one of the three time points) so that we have "biological" replicates?

i.e - Drug 1 - 3 barcoded samples (made from aggregate RNA from 9 total separate plates) from time point 1, 2, and 3... and same for Drug 2, 3, and 4?

Thanks!

Dave Brohawn