Low custom tdtomato gene content

Question

I have a set of scRNA-seq samples expressing TdTomato, which has high content in microscope. I followed the 10x cellranger pipelines to finsh the work, my procedures are as follows:

1. added TdTomato on an additional chromosome:
   

2. added an entry in the gtf

3. cellranger mkref
4. cellranger count

I got the result, which includes:

But strangely, the tdtomato expression content in my three sample are all very low! And cells with tdtomato are also very low and disperse!! Take sample 1 as example, you can see the picture:

It's not normal, but I can't find the reason. Following the 10x genomics help documentation noticed that by executing the CellRanger filtering steps, maybe my tdtomato reads are filtered too much, but because I'm not very skillful at programming and I'm first time to do RNA-seq (and scRNA-seq), I just don't know how to analysis the bam file as suggested by the official website and find the real reason of the problem:

If you can show me the code need to do, I will be very grateful!

Answer 1

A few suggestions:

1. Manually inspecting the BAM file with a genome browser like IGV, as @GWW suggested and checking the qualities of the alignments in your BAM/SAM file (SAM is the human readable version of BAM, there are tools to convert one to another);
2. Using another single cell analysis tool like Seurat so that you can manually set thresholds for QC and not lose to many (tdtomato) expressing cells at the QC step (I don't know about your experiment but if transfected, cells could be at a worse condition than non-transfected ones and hence might not pass QC);
3. Double checking your tdtomato reference sequence, I don't know about tdtomato but for example GFP has flavors like EGFP, similar but different sequences, might explain low alignment score if you observe in step 1;
4. Taking the limitations of scRNA-seq (especially droplet-based) technologies into account. The figure below is from the CITE-seq paper (from supp info), Simultaneous epitope and transcriptome measurement in single cells (Stoeckius et al., 2017), showing that the protein - RNA measurements of a given gene might show poor correlation.

Answer 2

The solution to your problem probably is adding the full mRNA sequence of your transgene to the reference (as also suggested by acrux).
I had a very similar problem recently when tdTomato expression was only detectable in a couple of cells. The transgene was introduced into the cells using a lentiviral vector and therefore the mRNA had the tdTomato sequence followed by a WPRE element. In the picture below you can see that a large number of reads is mapping to the WPRE element in the 3' UTR (blue region) and and almost none to the tdTomato CDS (red region).

In general, the untranslated regions are considered exons because they are part of the final mRNA. So if you have the fasta sequence beginning at the transcription start and ending at the polyadenylation signal, this is the full mRNA. Then you just need to update the GTF with the longer full mRNA length and you are all set.

Answer 3

We had the same issue. Turned out, our tdTomato transgene was followed up by WPRE and BGH polyA signals. These sequences are quite large, and their presence precludes sequencing into the tdTomato coding sequence when using the 10x genomics platform. Incorporating these into our build, allowed us to detect tdTomato+ cells.

Answer 4

I am unable to respond to other's comments as I never usually comment, so this is a new account. However, I spent the last couple days really diving into the BAM alignments when mapping scRNAseq reads to the human genome, but altering the GTF and FASTA using different annotations of my CAR construct (some very liberal allowing upstream and to the poly A tail sequence, and some less liberal)

MY MAIN CONCLUSION!:

DO NOT ADD THE BGH POLY A SIGNAL TO THE GENE ANNOTATION! it is highly homologous to the human beta globin subunit, and I am finding almost none of the reads actually mapping both to the adjacent CAR construct and the BGH POLY A subunit.

I am still figuring out how to handle the CD3zeta, CD28, and the OX40 in the CAR construct being multi-mapped (I will test just using a small portion of the CD28 transmembrane domain and then using just the chimeric antigen portion for mapping)

BUT, BIGEST TAKEAWAY, if you are going to attempt to add the poly A tail or any other genomic annotation not strictly relating to your gene of interest look at the bam and ensure that the reads mapped actually bridges into your gene. (as I also had residual genomic DNA from human beta globin upstream of my CAR construct that showed splicing over the CAR construct and aligning downstream of the CAR to the POLY A tail sequence)

-for viewing the BAM i am using IGVtools, and for the creation of the GTF and the FASTAs that I concatenated to the human reference genome I used tears [not a software package] ;)

Answer 5

you might want to try adding the 3'UTR sequence of you tdtomato construct used in your cells. that should do the trick.