No counts for added gene in cellranger (scRNA-seq)

Question

I have a set of scRNA-seq samples enriched with FACS for cells expressing a specific gene reporter (TdTomato). In particular the gene I want to report has positive counts in the resulting matrix for 97% of cells.

I followed CellRanger documentation to create a custom reference. I added TdTomato on an additional chromosome and I added an entry in the gtf:

TdTomato    .   exon    1   1431    .   -   .   gene_id "TdTomato"; transcript_id "TdTomato"; gene_name "TdTomato"

The command I'm using is:

cellranger-2.1.1/cellranger mkref --genome=refdata-cellranger-mm10-2.1.0.tdtomato --fasta=genome.tdtomato.fa --genes=genes.tdtomato.gtf

where genome.tdtomato.fa is the original reference genome (mm10, as provided by cellranger) with an additional chromosome at the end with header:

>TdTomato

and genes.tdtomato.gtf the original reference genome gtf with the additional line indicated previously at the end of the file.

I actually tried both on the '+' and on the '-' strand, however I guess this should not affect the read mapping, since TdTomato is on an independent chromosome.

I ran cellranger count using the custom reference:

s="SAMPLE"
cellranger-2.1.1/cellranger count --id="$s"_tt --transcriptome=refdata-cellranger-mm10-2.1.0.tdtomato --fastqs=original_fastqs_directory/"$s"/fastqs --sample="$s"

I am using the default parameters of CellRanger.

However, I found that less than 1% of cells seems to express TdTomato. If I use the '+' strand I get slightly more counts (2.4% of cells have positive TdTomato).

What I checked:

• With IGV I can see that there are reads mapping to TdTomato
• Using samtools I am able to see the same number of reads mapped on TdTom either with the + or - strand
• However, following this documentation I noticed that by executing the CellRanger filtering steps, more reads are filtered in the case of the - strand

This reduces the number of reads used for counts to 44 (2.4% of the original 1807 reads, - strand), 240 (13.3%, + strand) for TdTomato, 71620 (21.2%) for the gene reported by TdTomato. Interestingly, there are fewer reads for TdTom than counts, so there may be some kind of normalization afterwards?

The gene reported by TdTomato is longer, 11394 bp than TdTom, 1431 bp. The total length of the exons for the two genes is 5349 vs TdTom: 1431. Does this influence exponentially the amount of reads mapped to the two genes?

Answer 1

The first column of your GTF must correspond exactly to the name of the sequence in your FASTA file. Typically this is the chromosome so a GTF line:

chr1    hg38_knownCanonical exon    11106535    11262507    0.000000    .   .   gene_id "gene1"; transcript_id "tx1"; 

This must have a corresponding sequence in the FASTA file:

>chr1
ATCG ATCG ATCG ATCG ATCG ATCG...

Therefore you need to add an additional sequence to your FASTA file with >TdTomato. This must have exactly the same sequence as the construct that you are expecting reads for. The run cellranger mkref and cellranger count on these FASTA and GTF files. Cellranger also can take a GTF with “CDS” as a feature in addition to “exons”. The STAR alignment algorithm used by cellranger is splice aware so it can handle split reads but you will need to ensure the GTF corresponds to exons of the mRNA/cDNA rather than the full sequence.

10x Genomics is a platform with a high dropout rate, especially for lowly expressed genes. You should bear this in mind for what you expect to observe. It may be worth comparing the mapping rate of your construct to housekeeper genes to see if this is the reason for the low counts. Since whole mRNA molecules are captured on the beads before preparing the library, there should not be a difference in UMI counts between long and short genes.