Why full-length sequencing is not possible in UMI-based methods in single-cell RNA-seq

Question

[lsy9]

Why is full-length sequencing incompatible with UMI-based single-cell RNAseq methods, and all UMI-based methods (Cell-seq, Drop-seq, etc.) only sequence 3' or 5'-most fragments? Is it impossible fragment the mRNA/cDNA molecules and add an UMI to each fragment?

Answer 1

[i.sudbery]

Its entirely possible that one could create a full length single cell RNA seq protocol with UMIs, its just that nobody has done. There are bulk RNA-seq protocols with UMIs, such as Lexogen's CORALL kits.

[dsull]

I'm curious about how Lexogen would work because my understanding is that Illumina sequencing is limited by fragment size (i.e. fragments should be short and longer fragments = worst performance -- one reason that size selection is often performed). I haven't been able to find any well-done benchmarks on the quality of Lexogen libraries.

[i.sudbery]

The Lexogen kits are tagmentation based:

You do cDNA synthesis (both first and second strand), then you tagment - this uses a recombinase to "recombine" the cDNA with a a sequencing adaptor at a random location. Thus you fragment and tag (or add adaptors) at the same time. In the CORALL kits, that adaptor has a UMI sequence on it. You then PCR amplify.

Now, if you want to count transcripts with this, you are making the assumption that each cDNA will only be tagmented in one location, which is likely, but not guaranteed. There is no way to ensure that two different reads from different parts of the transcript do or do not come from the same original molecule. But what you can do is remove PCR bias - you can tell whether two reads that come from the same place in the transcript come from the same molecule or not.

This is good enough to ensure that counts between samples are truly proportional, but not to give absolute numbers (end-tagging doesn't give absolute numbers either, because it assumes an equal recovery efficiency across all genes, and that you know that recovery efficiency, neither of which are true).

[igor]

Doesn't Fluidigm C1 have a UMI option?

Answer 2

[dsull]

Well, if you attach a UMI to one end, you can't get full-length reads if you use a fragmentation-based protocols because, then, only one fragment originating from the same RNA molecule will have the UMI!

You can't attach a UMI to each fragment because you don't know what molecule that fragment originated from, following fragmentation.

There's smart-seq3 which can get you UMI reads and full-length reads separately.