Why Are Transcripts Differentially Expressed but Not Genes in RNA-seq Analysis? — Confusion Around Transcript-Level Differential Expression

Question

[Cooper]

Hello everyone, I'm facing a challenge while trying to conduct RNA-seq analysis. After completing the differential expression analysis using DESeq2, I found that there are very few differentially expressed genes, but there are quite a few differentially expressed transcripts. Why do transcripts show differential expression while genes don't?

I carefully checked the count data and noticed that some transcripts show differential expression, but other isoforms of the same gene exhibit an opposite trend, or their expression levels are much higher than the differentially expressed transcript. In this case, should I trust the results of the differentially expressed transcripts?

I'm planning to use another method (Htseq) for counting, and I'm not sure if the results will differ from those obtained using RSEM. Additionally, I checked these transcripts on Gene Browser and found that some of the differences are quite small. Does this really matter? Will the encoded protein be different?

In short, I'm very confused about how to interpret differential transcripts. If there are any good courses that can help me, I would greatly appreciate it!!

[Ram]

RSEM calls transcript level expression first and then aggregates them to gene-level based on the length and percentage of each transcript. I'm not sure if that helps but just thought you should have a bit more information if you don't know this already.

Answer 1

[i.sudbery]

Why might there be more DE transcripts than DE genes?

1. A gene has many transcripts, so each DE gene probably contributes many transcripts. A gene with 10 isoforms that is upregulated will cotribute upto 10 upregulated isoforms, as long as splicing doesn't change.

2. A gene might be differentially spliced. With differential splicing, you don't neccessarily get a change in gene level. So if a gene has two transcripts, one splicing in a cassette exon, and the other not, you might see the inclusion level of that exon change between conditions without a change in transcript rate. This would lead to one isoform being up regulated, and the other being down regulated, giving 2 DE transcripts, but not DE genes. However, it doesn't have to be like this - if one transcript is down significantly down regulated, the upregulation could be spread amoungst the other isoforms such that no one other isoform is upregulated enough to be significant.

3. Consider the following count table:

                | Condition A  (3 reps) |  Condition B ( 3 reps)
   -------------+-----------------------+------------------------
   Transcript A |         0,0,0         |      10, 10, 10
   Transcript B |     1000, 800, 1500   |     900, 1200, 1300
   Gene         |      1000, 800, 1500  |     910, 1210, 1310
   

Transcript A is clearly different between condition A and B, but transcript B is 800-1500 in condition A and 900-1300 in condition B, averageing out about th same, and the difference definationly being less than the variance. Because B is so much more highly expressed than A, then the same applies. While the difference is marginally higher for the gene than the transcript, this difference is still swamped by the variance.

In general, transcript quantificaiton is less accurate than gene quantification, in part because its often obvious which gene a read comes from, but not which transcript. Unless you are specifically interested in splicing changes, I'd ingore the transcripts DE, and only look at the gene DE. If you are interested in splicing differences, I'd use a pipeline taylored to that, such as DEXSeq, rMATs, DRIM-seq or Swish.

[Cooper]

Thank you so much for your clear explanation! This really helps me understand the difference between transcript and gene-level DE. I appreciate the tool recommendations and your time in helping me. Thanks again!