RNAseq design: differential expression between sexes in several species

Question

[anonyjooj]

Hi everybody,

I'm wondering if my unusual sequencing design seems ok to you.

I will sequence one male and one female (no biological replicates) in ~50 species of one animal taxa.

My only goal is to see general differential expression between the male condition and the female condition in this taxa, then compare the male/female difference between 2 groups of species (by example, 25 species with a strong sexual dimorphism vs 25 species without important sexual dimorphism).

I know that biological replicates are important, but here some species are difficult to obtain and I prefer to spend money on more species than more replicates of the same species. Here, the "biological replicates" would be the 50 males and the 50 females of different species. I know that it is unusual, but I believe that it can be an interesting approach to get a general pattern.

Does it seem a valid opinion to you?

Which strategy could be the best to achieve the analyses?

I though to do classical differential expression analyses without replicates on each species, then compare for each 1-1 ortholog genes the p-value of the 25 species (high sexual dimorphism) vs the p-values of the 25 other species (low sexual dimorphism).

Thanks for your feedback.

[Michael]

No replication is not a valid design for DE analysis.

Answer 1

[h.mon]

If you prefer to get a lot of unreliable data for a lot of species, then go ahead.

Be aware that in addition to poor data, you have to account for phylogenetic effects, so it would be useful if you describe your taxonomic sampling a bit better.

[anonyjooj]

I'm well aware of the trade-off and I know that I sacrifice reliability of the gene expression data for more genomic data on new non-model species (that will be useful for another purpose). My question is more if the DE results will be necessarily meaningless considering the fact that I will have multiple comparisons in a lot of species.

Be aware that in addition to poor data, you have to account for phylogenetic effects, so it would be useful if you describe your taxonomic sampling a bit better.

I'm well aware of the phylogenetic effects, it's more in my area of expertise :)

Answer 2

[Devon Ryan]

I'm in total agreement with h.mon, you'd be better served sequencing fewer species with more replicates than more species with fewer replicates. There's going to be a species-specific difference in males vs. females, at least to a certain extent, and without biological replicates that'll confound your results.

[anonyjooj]

Thanks for your feedback.

>you'd be better served sequencing fewer species with more replicates than more species with fewer replicates

Well, the problem is that I don't want to generate these data ONLY for differential expression, but also to get a big genomic dataset of new species for phylogenomics and detect positive selection (and I really need a lot of species for this aspect). I know that the design is not perfect for ONLY differential expression, but I would like to know if I can also exploit the expression data with this sequencing plan or if it's better to forget it.

>There's going to be a species-specific difference in males vs. females, at least to a certain extent, and without biological replicates that'll confound your results.

Does it mean that I'm likely to obtain no significant genes or tons of false positives? My idea was first to take into account the variation of the species-specific difference to get the genes that are really differently expressed between all males and all females of all the taxa. What would be the consequence to estimate the common dispersion between all males of the dataset? A dispersion too important (because of species-specific differences) and no significant genes?

Answer 3

[Nicolas Rosewick]

Be also aware to use the same cell type in every different species (not an easy job..)