Merging transcripts of two genotypes

[pravee1216]

I have one question related to constructing transcriptome using de novo of two genotypes for expression studies.

I have two conditions - treated and control samples - from two genotypes of a polyploid (poorly annotated) plant . I would like to do a comparative transcriptomics and identify differently regulated known/novel unigenes between two genotypes. Is it advisable to combine the unigenes from both genotypes and use this as reference for differential expression analysis and comparison between genotypes, as it is polyploidy? Or I should construct the unigenes at genotype level (by merging treated and control transcripts) and compare their expression between genotypes based on the annotation??

Please suggest and support with some references

Best
Raj

[keithforest]

From my experience, you may be best off combining all reads and making a single transcriptome assembly. Then mapping reads (preferably with replicates) back to the unified assembly would allow you to identify differentially expressed transcripts between conditions/genotypes.

As for references, I can't think of any that addressed this issue off the top of my head.

[pravee1216]

Thanks, but I recently came through an article: Transcriptome analysis of two buffalograss cultivars (M Wachholtz et al) discussed the same scenario. Here are some key text from this article:-

"Separate transcriptomes were assembled for each genotype. Due to the polyploid nature of these plants and a potential high level of intra-organism and inter-organism variation, such as genome rearrangements or paralogue genes unique to one genotype, we decided to not combine genotype reads. Combining reads from both genotypes could potentially complicate the assembly process and create inaccurate transcripts. Assembly was performed using Velvet/Oases software [43]. Multiple assemblies were created per genotype, using odd k-mer values 27-51. Previous studies have shown that using multiple assemblies, at varying k-mer values, captures more lowly expressed transcripts when compared with a single k-mer assembly [44]. Combined transcripts from the multiple k-mer assemblies were run through the CD-HIT-EST program to remove redundant transcripts sharing 100% identity [41]."

I'm wondering whether this is the case for polyploid species.

[chadn737]

If you have access to a server, you might want to try it both ways. What organism are you working on? I've messed around with enough polyploid data to know how painful it can be.

The optimal solution may depend on your experimental design. How many replicates do you have (I hope you have replicates) and how deep did you sequence? If you do each treatment or genotype separately and have insufficient sequencing, then you aren't going to reconstruct a lot transcripts, certainly not as many as you would by combining all the data. If you have sufficient replicates and sequencing, then that becomes less of an issue.

If you do it separately, then you also run into the problem of correctly comparing you assemblies for differential expression. For instance, are you sure that gene A in genotype A is the same as gene A in genotype B? This is especially true if in one genotype, due to higher expression and more reads, you have a longer transcript than in the other sample. Or if in one sample, the same transcript is actually multiple fragmented contigs. This could be more difficult and labor intensive than doing the de novo assembly itself. Where as if you do the assembly altogether and then realign to this, you at least are more confident that you are comparing the same thing.

[Jeremy]

I don't agree with the statement highlighted in Wachholtz et al., plus it is very vague: "could potentially complicate the assembly process" ? what does that mean? How?

Either way you will get a lot of junk assembled into transcripts. When you map the reads back to the transcript the junk will have very few reads mapped to it so you can clean up your assembly that way. Genes unique to one sample should not cause a problem. Merging the assemblies afterwards will run you right into the same problems as doing the assembly with the combined reads, the only difference is that it much more manual combining the reads later than just doing it automatically at the start.

E.g. You have a gene that has en extra exon in the middle of it in one sample but not the other: combining the reads at the start will give two isotigs from the one isogroup that you can directly compare. Combining the isotigs afterwards from separate assemblies will give you two different isotigs that may or may not cluster in CD-HIT that you then need to manually figure out is actually the same gene, but the extra exon might only change the read depth by 10-20% leaving it out of your list of DE genes. While from the combined assembly, as long as you use a program that can pick up DE alternate transcripts, you should find it.