crispr-offtarget.md

Overview

This guide is how to call offtarget edits in a CRISPR edited genome. This is pretty easy to and only takes a few steps. First, we need to figure out what is different between the CRISPR edited samples and the (hopefully they gave you these) control samples. Then we need to find a set of predicted off-target CRISPR sites. Finally, once we know what is different, we need to overlap the differences with predicted off-target sites, allowing some mismatches. Then we can report the overall differences and differences that could be due to offtarget edits.

Call CRISPR-edited specific variants

You want to call edits that are in the CRISPRed sample but not the unedited sample. You can do that by plugging into the tumor-normal calling part of bcbio and pretending the CRISPR-edited sample is a tumor sample and the non-edited sample is a normal sample.

To get tumor-normal calling to work you need to use a variant caller that can handle that, I recommend mutect2.

To tell bcbio that a pair of samples is a tumor-normal pair you need to

1. Put the tumor and normal sample in the same batch by setting batch in the metadata to the same batch.
2. Set phenotype of the CRISPR-edited sample to tumor.
3. Set the phenotype of the non-edited sample to normal.

And kick off the variant2 pipeline, the normal whole genome sequencing pipeline. An example YAML template is below:

---
details:
  - analysis: variant2
    genome_build: hg38
    algorithm:
      aligner: bwa
      variantcaller: mutect2
      tools_on: [gemini]

And an example metadata file:

samplename,description,batch,phenotype,sex,cas9,gRNA
Hs27_HSV1.cram,Hs27_HSV1,noCas9_nogRNA,normal,male,no,yes
Hs27_HSV1_Cas9.cram,Hs27_HSV1_Cas9,noCas9,normal,male,yes,no
Hs27_HSV1_UL30_5.cram,Hs27_HSV1_UL30_5,noCas9_nogRNA,tumor,male,yes,yes
Hs27_HSV1_UL30_5_repeat.cram,Hs27_HSV1_UL30_5_repeat,noCas9,tumor,male,yes,yes

Find predicted off-target sites

There are several tools to do this, a common one folks use is cas-offinder, so that is what we will use. There is a web app but it will only return 1,000 events per class. Usually this is fine, but if you allow bulges you can get a lot more offtarget sites so you might bump into this limit.

First install cas-offinder, there is a conda package so this is easy:

conda create -n crispr -c bioconda cas-offinder

There is a companion python wrapper cas-offinder-bulge that can also predict offtarget sites taking bulges into effect. You can download it here if you need to do that.

You'll need to know the sequence of one or more guides you want to check. You will also need to know the PAM sequence for the endonuclease that is being used.

You can run cas-offinder like this:

cas-offinder input.txt C output.txt

where input.txt has this format:

hg38.fa
NNNNNNNNNNNNNNNNNNNNNNNGRRT
ACACGTGAAAGACGGTGACGGNNGRRT 6

hg38.fa is the path to the FASTA file of the hg38 genome. NNNNNNNNNNNNNNNNNNNNNNNNGRRT is the length of the guide sequence you are interested in with the PAM sequence tacked on the end. ACACACGTGAAAGACGGTGACGGNNGRRT is the guide sequence with the PAM sequence tacked on the end. 6 is the number of mismatches you are allowing here, it will look for sites with that many or less mismatches.

If you want to look for bulges, use cas-offinder-bulge with this format:

hg38.fa
NNNNNNNNNNNNNNNNNNNNNNNGRRT 2 1
ACACGTGAAAGACGGTGACGGNNGRRT 6

Where the 2 says to look for a DNA bulge and the 1 a RNA bulge. You can do one or the other, neither or both.

After you run cas-offinder you can conver the output to a sorted BED file for use with intersecting the your variants:

cat output.txt | sed 1d | awk '{printf("%s\t%s\t%s\n",$4, $5-10,$5+10)}' | sort -V -k 1,1 -k2,2n  > output.bed

Overlap variants

Finally, use the BED file of predicted off-target sites to pull out possible off-target variant calls:

bedtools intersect -header -u -a noCas9_nogRNA-mutect2-annotated.vcf.gz -b output.bed

And you are done!

More tools

CRISPResso2